voice/twilio: a-leg place_call should fail fast on busy / no-answer / failed instead of waiting the full stream-connect timeout

[drewdrewthis]

What happens

With attach_stream="a-leg" (#762, PR #982), place_call() issues Calls.create and then waits on _stream_connected for timeout (default 120s). If the callee never answers — Twilio marks the call busy, no-answer, failed or canceled within seconds — the adapter still sits for the full timeout before _abandon_originated_call hangs up and raises TimeoutError.

Observed twice on 2026-09-09 while producing the proof: Twilio returned busy at +6s, the adapter raised at +120s.

Proposal

Fail fast as soon as Twilio reports a terminal non-answered status, and surface it as a distinct error (CallNotAnsweredError(status=...)) rather than TimeoutError. Two ways to learn the status, either is fine:

1. Pass status_callback=<public_base_url>/twilio/status with status_callback_event=["answered","completed"] on Calls.create and have the server set an event; the a-leg path already owns the tunnel + server.
2. Poll calls(sid).fetch().status every ~2s while waiting on _stream_connected.

Keep TimeoutError for the case where the call is in-progress but the stream never connects (tunnel/WS problem).

Same for the JS adapter (placeCall / _abandonOriginatedCall).

🤖 Generated with Claude Code

https://claude.ai/code/session_01Tc6K2QfkvcWiasy8ane6EG