fix: replace assert with runtime guard in tts_tool streaming playback

[JonthanaHanh]

Summary

Two assert statements in stream_tts_to_speaker() inner functions (_playback_worker and _enqueue_audio) would be stripped by python -O, silently removing invariant checks.

Changes

• tools/tts_tool.py:3597 — _playback_worker(): assert streamer is not None → if streamer is None: return
• tools/tts_tool.py:3704 — _enqueue_audio(): assert streamer is not None → if streamer is None: return

Why

assert is stripped by python -O. If a future refactor breaks the invariant that streamer is non-None when these functions run, the code would crash with a confusing AttributeError instead of a clear guard. The early-return pattern is consistent with the rest of the codebase.

[spfcraze]

This was generated by AI during triage.

Summary:
The new guard in _enqueue_audio returns before the function's existing try/except, so the failure this PR hardens against — streamer None under python -O — now drops a sentence with no log line where the pre-change code logged one.

Problems:

• The new if streamer is None: return in tools/tts_tool.py (replacing the assert at line 3704) sits above the try block whose except Exception at line 3707 logs Streaming TTS synthesis failed: %s (line 3708). The pre-change code, with the assert stripped under python -O, would reach streamer.stream(...) -> AttributeError -> that except -> the warning; the guard returns before that path.
• The diff adds no logging of its own (no logger token appears), so nothing records the dropped sentence — _speak_sentence has already displayed it and marked it spoken.

Solution:
Log in the guard before returning, matching the function's other failure paths: if streamer is None: logger.warning("Streaming TTS: streamer is None, dropping sentence"); return.

---

Checked against 8f03e89 — the PR head when this was written — and e5e96e8, main at the same moment.