fix(tts): transcode xai ogg output paths

[tianma-if]

Summary

• synthesize xAI TTS .ogg requests to a sibling MP3 first
• transcode that MP3 with the existing Opus helper so Telegram voice targets get real OGG/Opus
• return the actual synthesized path when conversion is unavailable
• add a regression test that keeps the xAI payload on the default MP3 codec while converting the local output

Fixes #57213

Tests

• .venv/bin/python -m pytest tests/tools/test_tts_xai_speech_tags.py -q -k "transcodes_explicit_ogg_path or omits_speed or leaves_text"
• .venv/bin/python -m ruff check tools/tts_tool.py tests/tools/test_tts_xai_speech_tags.py
• .venv/bin/python -m pytest tests/tools/test_tts_xai_speech_tags.py -q
• scripts/run_tests.sh tests/tools/test_tts_xai_speech_tags.py -q

[alt-glitch]

This was generated by AI during triage.

Duplicate of #57222 — both PRs fix #57213 by editing tools/tts_tool.py::_generate_xai_tts to synthesize xAI TTS to a sibling MP3 and transcode it to Opus via _convert_to_opus for .ogg targets (same file, same mechanism, near-identical regression test). #57222 is the earlier open PR, kept as canonical. Both sit in the broader .ogg/Opus TTS cluster (#54589 / #54597 / #57190).

[tianma-if]

Closing as duplicate of #57222 per triage; the earlier PR covers the same xAI TTS OGG transcoding fix. Thanks.