fix(gateway): request ogg for telegram auto tts

[tianma-if]

Summary

• carry the destination platform explicitly into both gateway auto-TTS paths, including after inbound session ContextVars have been cleared
• keep conversion-backed providers on a real source path (Edge writes MP3), then return the converted OGG/Opus path for Telegram
• let native-Opus providers select .ogg from delivery intent rather than relying on a caller-provided suffix
• keep non-Telegram delivery on the existing MP3 behavior

Verification

• pytest -q tests/tools/test_tts_*.py tests/gateway/test_auto_voice_reply_format.py tests/gateway/test_base_topic_sessions.py (262 passed)
• Ruff passes for all modified implementation and test files

[teknium1]

Thanks for tracing the adapter-level path; the missing context after the runner returns is real. However, the proposed filename change is not sufficient for conversion-backed providers.

Problems

• gateway/platforms/base.py:4978 passes a .ogg path to text_to_speech_tool, but Edge writes MP3 output (tools/tts_tool.py:947-971). Opus conversion only runs when want_opus is true and the produced path is not already .ogg (tools/tts_tool.py:2562-2570). The adapter call occurs after GatewayRunner clears the session context (gateway/run.py:12448-12450), so this path has want_opus == false; the result can be MP3 bytes under an .ogg name.
• The test only checks the mocked requested suffix, so it cannot prove conversion or valid Opus delivery.
• gateway/run.py:13302-13310 has the same output-path pattern and should be handled with the same corrected mechanism.

Suggested changes

• Carry explicit Telegram/Opus intent into the TTS layer and preserve a real conversion input for Edge/WAV-backed providers; cover the converted result, not only the requested filename.
• Apply that mechanism to both auto-TTS call paths.

Automated hermes-sweeper review.

[teknium1]

A .ogg suffix alone is not an Opus payload. Edge writes MP3 output, and text_to_speech_tool only calls _convert_to_opus when want_opus is true and the generated path does not already end in .ogg (tools/tts_tool.py:2562-2570). Here the runner has already cleared the session context, so this can send MP3 bytes under an .ogg name. Please pass explicit delivery intent into TTS and preserve a real conversion input instead.

[teknium1]

The root cause here (gateway auto-TTS choosing MP3 vs Ogg/Opus via the cleared HERMES_SESSION_PLATFORM contextvar, so opus platforms got audio attachments instead of native voice bubbles) was fixed class-wide in #73508: platform-awareness now comes from the caller via build_auto_tts_output_path(platform) keyed off OPUS_VOICE_PLATFORMS (based on @giladbau's #62040), with the central container repair guaranteeing real Ogg/Opus bytes.

(Landed via #73508, merge f440a44753.) Closing.