Problem

When a transient DNS failure causes the Slack socket mode WebSocket to close, SocketModeClient calls delayReconnectAttempt(this.start) to reconnect. Internally, the WebClient retries apps.connections.open up to 100 times (hence the 2 800+ retry count in the report). If those retries are exhausted before DNS recovers, the resulting RequestError is swallowed (unhandled rejection inside delayReconnectAttempt) and:

• State.Disconnected is never emitted
• No new State.Connected event arrives
• The socket is permanently dead; the gateway requires a manual restart

This is the same class of bug as #13506 (WhatsApp).

Root cause

WebSocket closes
  → SocketModeClient.on(close) fires
  → delayReconnectAttempt(this.start) is called
    → this.start() → retrieveWSSURL() → apps.connections.open() (100 HTTP retries)
    → all retries fail (ENOTFOUND) → ErrorCode.RequestError → isRecoverable = false → throws
  → error propagates into delayReconnectAttempt().then(res) with no .catch()
  → unhandled rejection — socket silently dead forever

Fix

After a successful app.start() in socket mode, attach listeners to the underlying SocketModeClient (accessed via the internal receiver) for its "close" and "connected" events:

• On "close" — start a 3-minute watchdog timer (long enough for the SDK's own HTTP retry loop to succeed during normal transient outages).
• On "connected" — clear the timer (SDK reconnected on its own; nothing to do).
• If the timer fires — the SDK gave up silently; call app.stop() then app.start() again on the same App instance (same event handlers, new WebSocket) and back off exponentially (1 s → 30 s) between attempts.

HTTP mode is unaffected.

Implementation notes

• SocketModeClient emits "close" whenever the underlying SlackWebSocket disconnects (pong timeout, server close frame, etc.) — this is the reliable hook point.
• SocketModeClient.start() resets this.shuttingDown = false and creates a fresh SlackWebSocket, so calling it again after stop() is safe.
• The 3-minute watchdog is intentionally generous: SocketModeClient uses retryConfig: { retries: 100, factor: 1.3 } for apps.connections.open, so a short watchdog would race against the SDK's own recovery.
• Listeners are removed in the finally block of each iteration to prevent duplicates across reconnect cycles.

Fixes #27077