feat(telegram): split connector into home validator + remote bot

Move all Telegram egress off the main host. The single connector held the
bot token, long-polled Telegram and answered the gateway/backend over the
trusted internal network, so the whole component (including login validation)
shared fate with its VPN sidecar. Split it into two binaries that share the
token:

- cmd/validator (home, no VPN): Mini App initData + Login Widget HMAC only,
  never calls the Bot API. The gateway dials it for Telegram auth, so game
  login is now independent of Telegram reachability.
- cmd/bot (remote): Bot API long-poll + sendMessage, the only component
  reaching Telegram. It holds no inbound port — it dials the gateway over a
  new reverse mTLS bot-link (pkg/proto/botlink/v1) and executes the send
  commands the gateway pushes.

The gateway funnels sends to the bot-link: out-of-app push is fire-and-forget
(at-most-once, dropped if no bot is connected); the backend admin broadcasts
reach a gateway-served relay that forwards them and awaits the bot's ack
(SendToUser/SendToGameChannel contract preserved). mTLS (pkg/mtls) is the one
inter-service link that leaves the trusted segment; validator<->gateway and
the relay stay plaintext internal. The bot is Telegram-rate-limited.

One bot now; the gateway bot registry, an owns_updates flag and per-command
ids leave seams for N later. Webhook rejected (one URL per token, adds inbound
+ a static address).

The unified test contour runs the split (the bot keeps its VPN sidecar and
dials the gateway by its internal name; bot-link certs from deploy/gen-certs.sh,
generated in CI). The prod wiring — the bot on a separate host (no VPN), the
gateway bot-link port published, PROD_ certs with scheduled rotation, an SSH
deploy of both hosts together — is the deferred final stage (PRERELEASE.md TX,
Stage 18).

Docs: ARCHITECTURE, PRERELEASE (phase TX), platform/telegram + gateway +
backend + deploy READMEs, FUNCTIONAL(+ru), CLAUDE.md, .env.example.

gateway/cmd/gateway/main.go

@@ -9,16 +9,23 @@ package main
 import (
 	"context"
 	"errors"
+	"fmt"
 	"log"
+	"net"
 	"net/http"
 	"os/signal"
 	"syscall"
 	"time"
 
+	"go.opentelemetry.io/contrib/instrumentation/google.golang.org/grpc/otelgrpc"
 	"go.uber.org/zap"
+	"google.golang.org/grpc"
+	"google.golang.org/grpc/credentials"
+	"google.golang.org/grpc/keepalive"
 
 	"scrabble/gateway/internal/admin"
 	"scrabble/gateway/internal/backendclient"
+	"scrabble/gateway/internal/botlink"
 	"scrabble/gateway/internal/config"
 	"scrabble/gateway/internal/connector"
 	"scrabble/gateway/internal/connectsrv"
@@ -26,9 +33,23 @@ import (
 	"scrabble/gateway/internal/ratelimit"
 	"scrabble/gateway/internal/session"
 	"scrabble/gateway/internal/transcode"
+	"scrabble/pkg/mtls"
+	botlinkv1 "scrabble/pkg/proto/botlink/v1"
+	telegramv1 "scrabble/pkg/proto/telegram/v1"
 	pkgtel "scrabble/pkg/telemetry"
 )
 
+const (
+	// botLinkKeepaliveTime is how often the gateway pings an idle bot-link stream
+	// to hold the WAN connection open and detect a dead bot.
+	botLinkKeepaliveTime = 30 * time.Second
+	// botLinkKeepaliveTimeout bounds the wait for a keepalive ping reply.
+	botLinkKeepaliveTimeout = 10 * time.Second
+	// botLinkMinPingInterval is the smallest client ping interval the gateway
+	// tolerates before treating it as abuse.
+	botLinkMinPingInterval = 10 * time.Second
+)
+
 const (
 	// shutdownTimeout bounds the graceful HTTP shutdown.
 	shutdownTimeout = 10 * time.Second
@@ -100,17 +121,47 @@ func run(ctx context.Context, cfg config.Config, logger *zap.Logger) error {
 	tracker := ratelimit.NewTracker()
 	hub := push.NewHub(0)
 
-	var conn *connector.Client
 	var validator transcode.TelegramValidator
-	if cfg.ConnectorAddr != "" {
-		conn, err = connector.New(cfg.ConnectorAddr)
-		if err != nil {
-			return err
+	if cfg.ValidatorAddr != "" {
+		conn, cerr := connector.New(cfg.ValidatorAddr)
+		if cerr != nil {
+			return cerr
 		}
 		defer func() { _ = conn.Close() }()
 		validator = conn
 	} else {
-		logger.Warn("telegram disabled (GATEWAY_CONNECTOR_ADDR unset)")
+		logger.Warn("telegram auth disabled (GATEWAY_VALIDATOR_ADDR unset)")
+	}
+
+	// The reverse bot-link: the remote Telegram bot dials this gateway over mTLS and
+	// the gateway pushes send commands down the stream. Out-of-app push is
+	// fire-and-forget; the backend admin relay (plaintext, internal) awaits the Ack.
+	var botHub *botlink.Hub
+	if cfg.BotLinkEnabled() {
+		botHub = botlink.NewHub(logger, tel.MeterProvider().Meter("scrabble/gateway/botlink"))
+		tlsCfg, terr := mtls.ServerConfig(cfg.BotLink.CertFile, cfg.BotLink.KeyFile, cfg.BotLink.CAFile)
+		if terr != nil {
+			return terr
+		}
+		botSrv := grpc.NewServer(
+			grpc.Creds(credentials.NewTLS(tlsCfg)),
+			grpc.StatsHandler(otelgrpc.NewServerHandler()),
+			grpc.KeepaliveParams(keepalive.ServerParameters{Time: botLinkKeepaliveTime, Timeout: botLinkKeepaliveTimeout}),
+			grpc.KeepaliveEnforcementPolicy(keepalive.EnforcementPolicy{MinTime: botLinkMinPingInterval, PermitWithoutStream: true}),
+		)
+		botlinkv1.RegisterBotLinkServer(botSrv, botHub)
+		if serr := serveGRPC(ctx, "botlink", cfg.BotLink.Addr, botSrv, logger); serr != nil {
+			return serr
+		}
+		if cfg.BotLink.RelayAddr != "" {
+			relaySrv := grpc.NewServer(grpc.StatsHandler(otelgrpc.NewServerHandler()))
+			telegramv1.RegisterTelegramServer(relaySrv, botlink.NewRelayServer(botHub, cfg.BotLink.SendTimeout))
+			if serr := serveGRPC(ctx, "botlink-relay", cfg.BotLink.RelayAddr, relaySrv, logger); serr != nil {
+				return serr
+			}
+		}
+	} else {
+		logger.Warn("telegram bot channel disabled (GATEWAY_BOTLINK_ADDR unset)")
 	}
 
 	// The admin console (backend /_gm) is fronted on the public listener behind
@@ -142,8 +193,8 @@ func run(ctx context.Context, cfg config.Config, logger *zap.Logger) error {
 	})
 
 	// Bridge the backend push stream into the fan-out hub (and the out-of-app
-	// channel via the connector).
-	go runPushPump(ctx, backend, hub, conn, logger)
+	// channel via the bot-link).
+	go runPushPump(ctx, backend, hub, botHub, logger)
 	// Periodically summarise rate-limiter rejections (Warn log + backend report).
 	go runThrottleReporter(ctx, tracker, backend, logger)
 
@@ -195,6 +246,27 @@ func runServers(ctx context.Context, cancel context.CancelFunc, servers []*named
 	return first
 }
 
+// serveGRPC starts a gRPC server on addr in the background and gracefully stops it
+// when ctx is cancelled. It returns synchronously on a bind error so a
+// misconfigured listener fails startup fast; a later Serve error is logged.
+func serveGRPC(ctx context.Context, name, addr string, srv *grpc.Server, logger *zap.Logger) error {
+	lis, err := net.Listen("tcp", addr)
+	if err != nil {
+		return fmt.Errorf("gateway: listen %s (%s): %w", addr, name, err)
+	}
+	go func() {
+		<-ctx.Done()
+		srv.GracefulStop()
+	}()
+	go func() {
+		logger.Info("listener starting", zap.String("server", name), zap.String("addr", addr))
+		if err := srv.Serve(lis); err != nil && !errors.Is(err, grpc.ErrServerStopped) {
+			logger.Error("grpc listener failed", zap.String("server", name), zap.Error(err))
+		}
+	}()
+	return nil
+}
+
 // runThrottleReporter drains the rate-limiter rejection tracker on a fixed
 // cadence, emits one Warn summary per throttled key and forwards the report to
 // the backend (which feeds the admin throttled view and the high-rate
@@ -230,7 +302,7 @@ func runThrottleReporter(ctx context.Context, tracker *ratelimit.Tracker, backen
 // the hub and re-subscribing after the stream ends, until the context is done. For
 // the out-of-app push kinds it also routes events whose recipient has no live
 // in-app stream to the platform connector (a nil connector disables that channel).
-func runPushPump(ctx context.Context, backend *backendclient.Client, hub *push.Hub, conn *connector.Client, logger *zap.Logger) {
+func runPushPump(ctx context.Context, backend *backendclient.Client, hub *push.Hub, bot *botlink.Hub, logger *zap.Logger) {
 	for ctx.Err() == nil {
 		stream, err := backend.SubscribePush(ctx, gatewayID)
 		if err != nil {
@@ -255,10 +327,10 @@ func runPushPump(ctx context.Context, backend *backendclient.Client, hub *push.H
 				EventID: ev.GetEventId(),
 			})
 			// Out-of-app fallback: when the recipient has no live in-app stream,
-			// deliver the event over the platform push channel. Done in a goroutine
-			// so a slow connector never stalls the in-app firehose.
-			if conn != nil && connector.OutOfAppKind(ev.GetKind()) && !hub.HasSubscribers(ev.GetUserId()) {
-				go deliverOutOfApp(ctx, backend, conn, ev.GetUserId(), ev.GetKind(), ev.GetPayload(), logger)
+			// deliver the event over the bot-link. Done in a goroutine so a slow
+			// target lookup never stalls the in-app firehose.
+			if bot != nil && connector.OutOfAppKind(ev.GetKind()) && !hub.HasSubscribers(ev.GetUserId()) {
+				go deliverOutOfApp(ctx, backend, bot, ev.GetUserId(), ev.GetKind(), ev.GetPayload(), logger)
 			}
 		}
 		if !sleep(ctx, pushReconnectDelay) {
@@ -271,7 +343,7 @@ func runPushPump(ctx context.Context, backend *backendclient.Client, hub *push.H
 // Telegram identity and have not confined notifications to the app, asks the
 // connector to deliver the event. It is best-effort: every failure is logged and
 // dropped (the in-app stream remains the primary channel).
-func deliverOutOfApp(ctx context.Context, backend *backendclient.Client, conn *connector.Client, userID, kind string, payload []byte, logger *zap.Logger) {
+func deliverOutOfApp(ctx context.Context, backend *backendclient.Client, bot *botlink.Hub, userID, kind string, payload []byte, logger *zap.Logger) {
 	target, err := backend.PushTarget(ctx, userID)
 	if err != nil {
 		logger.Warn("push target lookup failed", zap.String("user_id", userID), zap.Error(err))
@@ -280,10 +352,9 @@ func deliverOutOfApp(ctx context.Context, backend *backendclient.Client, conn *c
 	if !connector.DeliverToTarget(target.ExternalID, target.NotificationsInAppOnly) {
 		return
 	}
-	// The single bot renders the message in the recipient's interface language.
-	if _, err := conn.Notify(ctx, target.ExternalID, kind, payload, target.Language); err != nil {
-		logger.Warn("out-of-app notify failed", zap.String("kind", kind), zap.Error(err))
-	}
+	// Fire-and-forget down the bot-link; the bot renders the message in the
+	// recipient's interface language and is dropped (best-effort) if no bot is up.
+	bot.Send(botlink.NotifyCommand(target.ExternalID, kind, payload, target.Language))
 }
 
 // sleep waits for d or until ctx is cancelled, reporting whether it waited the