fix(relay): prevent agent session corruption on timeout

[pinximei]

Summary

• fix(relay): After a relay timeout, the target agent became permanently unresponsive. Three root causes fixed:
  1. StartSession received the relay timeout context - when the deadline fired, exec.CommandContext killed the Claude process mid-turn, corrupting session state. Now uses the engine context.
  2. defer agentSession.Close() killed the process immediately on return. Now drains in a background goroutine (drainRelaySession) so the agent finishes its turn and saves state cleanly.
  3. No fallback when resuming a corrupted session - added retry logic that clears the stale session ID and starts fresh.
• fix(feishu): Respond to @ALL messages in group chats. Feishu @ALL sends @_all with 0 mentions, which was ignored when group_reply_all=false. Added respond_to_at_everyone_and_here option.

Test plan

• TestHandleRelay_ReturnsPartialOnTimeout - partial response returned, session closed cleanly in background
• TestHandleRelay_TimeoutWithoutTextReturnsContextError - error returned, no goroutine leak
• TestHandleRelay_ResumeFailureFallsBackToFreshSession - stale session ID cleared, fresh session created
• All existing relay tests pass

[chenhg5]

LGTM. Well-designed fix for relay timeout handling.

Review summary:

• ✅ Correct: Uses engine context instead of relay timeout context to prevent killing agent process
• ✅ drainRelaySession goroutine for clean session state preservation
• ✅ Fallback to fresh session on resume failure
• ✅ All tests pass
• ✅ CI passes

Good fix for session corruption issue.