fix(events): SSE health-check watchdog to detect silently-dead streams

[avfirsov]

Problem

The SSE for-await loop cannot detect a stream that died without throwing — it sits idle waiting for events that never arrive, and the reconnect logic only fires when the stream explicitly ends or errors. A corporate proxy (e.g. cntlm) cutting off long-running connections, a broken pipe that never surfaces an error, or a tight reconnect loop with no successful event delivery can all leave the bot silent indefinitely.

Changes

src/opencode/events.ts — track stream health and expose getters:

• lastSseEventTime is refreshed on every received event;
• consecutiveReconnectAttempts is incremented in both reconnect paths (stream-ended and error) and reset whenever an event arrives;
• exports getLastSseEventTime, getConsecutiveReconnectAttempts, isEventListening, getActiveEventDirectory.

src/bot/index.ts — add a 30-second sseWatchdogTimer:

• if isEventListening() is true AND we have not seen an event for >30s OR have piled up ≥5 reconnect attempts → stopEventListening() + ensureEventSubscription(directory);
• cleared in both createBot() and cleanupBotRuntime() alongside the existing heartbeat timer.

Test plan

• npm run build clean (verified locally)
• Start bot, normal traffic → watchdog stays quiet (event timestamps keep refreshing)
• Kill opencode serve while bot is connected → reconnects pile up → watchdog logs [SSE Watchdog] Restarting… → on opencode serve restart, subscription resumes without bot restart