fix: enable Opus voice output for Feishu and Lark platforms

[leowangzi]

Problem

TTS voice messages on Feishu/Lark are sent as file attachments instead of playable voice bubbles.

Root Cause

In tools/tts_tool.py, want_opus (which triggers ffmpeg conversion from MP3 to Opus and marks the output as voice_compatible) only checked for platform == "telegram". Feishu and Lark were excluded.

Meanwhile, the Feishu adapter only uploads .ogg/.opus files with the native "audio" message type (_FEISHU_OPUS_UPLOAD_EXTENSIONS = {".ogg", ".opus"}). MP3 files fall through to generic file attachment delivery.

Fix

One-line change: extend the want_opus condition to include feishu and lark:

- want_opus = (platform == "telegram")
+ want_opus = (platform in {"telegram", "feishu", "lark"})

This triggers:

1. ffmpeg MP3→Opus conversion for Edge TTS on Feishu/Lark
2. voice_compatible = True → [[audio_as_voice]] marker prepended
3. Opus file recognized by Feishu adapter → sent as native audio message

Testing

Tested on Feishu with Edge TTS: voice messages now arrive as playable audio bubbles instead of file attachments.

[alt-glitch]

This was generated by AI during triage.

Related: competing fix cluster for #45557. The earliest want_opus PRs are now closed (#45555, #45637, #32258); still-open alternatives are #45712 (Feishu-only want_opus, same one-line approach) and #37804 (combined gateway-side approach with duration metadata). This PR is the broadest of the live options — it extends the guard to both feishu and lark. Flagging for a maintainer to pick one.

[teknium1]

Thanks for narrowing the conversion gate to the Feishu/Lark delivery path. The current-main premise is valid: tools/tts_tool.py:2202 still restricts want_opus to Telegram, while the Feishu plugin routes .ogg/.opus as native audio at plugins/platforms/feishu/adapter.py:5025-5026.

Problems

• Please add a regression test for HERMES_SESSION_PLATFORM="feishu". tests/tools/test_tts_opus_routing.py:47-70 currently exercises only Telegram, so this new routing branch is unprotected.
• lark is not a current session-platform value: the Feishu/Lark adapter is constructed as Platform.FEISHU (plugins/platforms/feishu/adapter.py:1442) and the gateway exports that value (gateway/run.py:15000-15002). The feishu case covers Lark-domain deployments.
• This remains conversion-dependent: _convert_to_opus() returns None when ffmpeg is absent or fails (tools/tts_tool.py:899-933), after which non-Opus Feishu audio falls back to file routing (plugins/platforms/feishu/adapter.py:5034-5037).

Suggested changes

• Add the Feishu routing regression and limit the condition to the reachable platform identity.
• State the ffmpeg-success condition, or separately validate a fallback before promising native bubbles without it.

Automated hermes-sweeper review.

[teknium1]

Please add a Feishu regression in tests/tools/test_tts_opus_routing.py. The current test covers only Telegram; exercise this new branch with HERMES_SESSION_PLATFORM="feishu" and assert .ogg conversion plus the voice directive.

[teknium1]

lark is not emitted by the current adapter: Feishu/Lark constructs Platform.FEISHU, and the gateway exports its value as feishu. The feishu case already covers Lark-domain deployments; retaining an unreachable alias makes the routing contract misleading.

[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.