Subagent liveness/heartbeat: Task-spawned agents can die silently with no signal

Problem

Task-spawned subagents (including Codex-style detached workers) can die mid-execution and the orchestrator has no native way to know. The harness reports them as "running" indefinitely; the user only finds out by manually inspecting the process tree or noticing missing output hours later.

Concrete incident

• A Codex review subagent (pid 13224) died mid-review before the result harvest stage. The orchestrating session showed it as in-flight for ~1.5 hours before the user noticed. No error, no exit signal surfaced.
• This forced the creation of a user-land subagent-heartbeat convention: every dispatched subagent appends a timestamp every ~5 min to .claude/agent-status/<id>.log; the orchestrator polls every 10 min and treats >30 min idle as stalled.

That convention works but lives entirely in user-land — it requires every subagent author to opt in and every orchestrator to remember to poll.

Proposed

Surface subagent liveness in the harness:

1. Track last-output timestamp per Task agent.
2. Emit a SubagentStalled event (or set a status flag) when a subagent has produced no output for a configurable window (default ~30 min).
3. Optionally expose this in TaskList / TaskGet output so orchestrators can detect it without polling logs.

Use case

Long-running parallel agent dispatches (waves, backlog burndowns, codex reviews) where one silent death currently blocks the whole pipeline.