fix: auto-recover PTY sessions after unexpected exit (WiFi drops)

[wat-hiroaki]

Summary

Fixes PTY session recovery after WiFi disconnection or network errors.

Problem: When WiFi drops, Claude CLI exits with a network error. The session goes to error state and doesn't recover — even though claudeSessionId is preserved in the database. On manual restart, the conversation history was lost because --session-id doesn't restore context.

Root cause: Two issues:

1. startSession always used --session-id (starts fresh) instead of --resume (restores conversation)
2. No automatic retry logic on unexpected exit

Changes:

• --resume for existing sessions: When agent.claudeSessionId already exists, use --resume <sessionId> instead of --session-id <sessionId>. This tells Claude CLI to restore the full conversation history.
• Auto-recovery on unexpected exit: Up to 3 retry attempts with exponential backoff (2s → 4s → 8s). Recovery counter resets once the session becomes stable (active/thinking/tool_running).
• Safety guards: Skips retry if user explicitly stopped the session (isKilled). Re-reads agent from DB before retry to avoid stale state.

Test plan

• npm run lint — 0 errors
• npx tsc --noEmit — clean
• npm run build — clean
• npm test — 13/13 pass
• npm run dev — app starts without errors
• QA: Start a session → disconnect WiFi → reconnect → verify auto-recovery with conversation history
• QA: Verify max 3 retries (no infinite loop)
• QA: Verify user-initiated stop doesn't trigger auto-recovery
• QA: Verify fresh sessions (no prior sessionId) still use --session-id

Behavior summary

Scenario	Before	After
WiFi drop → reconnect	Session dies, stays error	Auto-retries up to 3x with backoff
Manual restart after crash	New session (conversation lost)	--resume restores conversation
User clicks Stop	Goes to idle	Goes to idle (no retry)
3+ consecutive failures	N/A	Stays error, no more retries

🤖 Generated with Claude Code