Add hard timeouts for stuck runtime requests

[gaelic-ghost]

What to add

Add hard timeout handling for stuck runtime requests so generation, playback, queue bookkeeping, and request event streams cannot remain active forever without a terminal state.

Why

A live service request can currently appear to remain running / playing after generated audio has completed. In the observed case, the worker stayed ready and generated audio was retained, but playback still reported an active request and DELETE /requests/{request_id} timed out.

Hard timeouts would keep the service operable when a backend, playback sink, stream, event continuation, or cancellation path fails to finish cleanly.

Suggested behavior

• Apply bounded timeouts around active generation requests.
• Apply bounded timeouts around active playback requests.
• Apply bounded timeouts around request cancellation so cancel routes return promptly.
• Emit a terminal event when a timeout fires.
• Include a human-readable timeout message that names the request ID, operation, profile name when available, elapsed time, and likely stuck stage.
• Clear or quarantine the active queue slot so later requests are not blocked by stale active state.
• Preserve enough request history to debug what stage timed out.

Related observed failure

• Request ID: 0217E899-B3D0-49A1-B1E0-55941BED6149
• Operation: generate_speech
• Voice profile: swift-signal
• Generated audio completed, but playback stayed active.
• Cancel route timed out and the request remained running.