[FEATURE]: Runtime watchdog for stuck tool/session recovery (complements #19023 startup recovery) #20099

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

This is a follow-up to #19023 (startup orphaned state recovery).

#19023 addresses cleanup at process startup after restart/crash. This request is specifically for a runtime watchdog while the server stays alive.

Requested runtime enhancements:

1. Periodic watchdog tick that scans for tool parts stuck in running longer than a configured threshold.
2. Leaf-level filtering so task tools waiting on child sessions are not force-failed prematurely.
3. Session-level idle detection with configurable threshold to cancel inactive stuck sessions.
4. Startup orphan cleanup remains covered by Sessions permanently stuck after server restart or stream interruption — no startup recovery for orphaned messages/tool parts #19023; this issue is intentionally scoped to runtime watchdog behavior.

Suggested config keys:

• experimental.tool_timeout
• experimental.task_timeout
• experimental.idle_timeout

Why this matters

Startup-only recovery does not help when a session wedges during normal runtime. If abort propagation fails or a child tool deadlocks, sessions can remain stuck indefinitely until manual intervention.

A runtime watchdog provides a safety net and keeps sessions progressing or failing fast with actionable errors.

Related

• Sessions permanently stuck after server restart or stream interruption — no startup recovery for orphaned messages/tool parts #19023 (startup recovery; this issue complements it)
• [BUG] Stream iteration hangs on abort; cleanup failure leaves zombie sessions #20094 (abort/cleanup deadlock path)