snippets: thread-queue-timer strands the second job's result and polls forever when a job raises #463

Why it matters

thread-queue-timer is the pattern the new timers skill (#446) recommends for long-running work. As written, results get stranded and a failing job leaves a timer polling forever. Agents will copy it verbatim.

Evidence (code reading)

• snippets/thread-queue-timer.py:29-48; the same pattern appears in skills/timers-modal-and-threading/SKILL.md:64-73.
• Stranded results: two run_in_background() calls start two workers. Once the first result arrives, _drain returns None, which unregisters the timer, and the second result sits in the queue until something registers a timer again.
• Endless polling: if job() raises, the worker thread dies without calling put(), and _drain re-runs every 0.1 s forever. Nothing reports the error.

Suggested approach

• Keep a pending-job counter, or one timer per job. Drain everything available in each tick, and unregister only when the pending count reaches 0.
• In the worker, wrap job() in try/except BaseException as e and put((False, e)). The drain re-raises or reports the error on the main thread.
• Extend examples/timers-modal-threading with a two-job case and a raising job, each with a falsifier.

Done when

The snippet and skill handle N concurrent jobs and job exceptions, and the example proves both on 5.2 and 4.5.