feat(gateway/stt): reply to voice messages with transcription text

[Quarkex]

Summary

Add an opt-in stt_reply_transcript config option that replies to voice messages with their transcription text, making voice content searchable and quotable in chat history.

Motivation

Voice messages in messaging platforms are opaque — you can't search them, quote them, or skim them. When STT is already transcribing for the agent, surfacing that transcription as a visible reply costs nothing extra and significantly improves UX for users who rely heavily on voice input.

Configuration

stt:
  reply_transcript: true  # default: false

Or flat: stt_reply_transcript: true

Implementation

• gateway/config.py: New stt_reply_transcript field with from_dict/to_dict support (accepts both nested stt.reply_transcript and flat key)
• gateway/run.py:
  • Extended _enrich_message_with_transcription signature to accept platform/chat_id/message_id/thread_id
  • New _send_reply_transcript helper — fire-and-forget, calls the platform adapter's send_message with reply_to_message_id
• tests/gateway/test_stt_reply_transcript.py: Config parsing tests (5 cases)

Design decisions

• Opt-in — disabled by default, zero behavior change for existing users
• Fire-and-forget — reply delivery failures are logged at DEBUG, never block message processing
• Not injected into agent transcript — the reply is purely a UX convenience; the agent sees only the normal [The user sent a voice message~] envelope
• Platform-agnostic — works with any adapter that implements send_message(reply_to_message_id=...)
• Reply format: 🎤 «transcribed text» — visually distinct, not confused with agent responses

Testing

• All existing STT tests pass (test_stt_config.py: 5/5)
• New test file: 5/5 pass
• Manually tested on Telegram (voice → reply appears instantly)

[alt-glitch]

Duplicate of #13727. This is the fourth independent implementation of STT transcript echo (after #13727, #22123, and #25196). The canonical PR is #13727 which adds stt.send_transcription config key.

[teknium1]

This is an automated hermes-sweeper review: the configurable STT transcript echo requested here is already implemented on current main.

• gateway/run.py:10435-10456 transcribes inbound audio and sends each successful 🎙️ transcript through the source platform adapter.
• gateway/config.py:689, 861-867, and 924 provide the canonical stt.echo_transcripts configuration gate; tests/gateway/test_stt_transcript_echo_config.py:8-70 covers its behavior and all echo send sites.
• gateway/run.py:15203-15224, 19180-19197, and 19602-19618 cover queued and active-run voice-message paths as well.
• The implementation shipped in bfc5262725a3a12b0f1c6d4b9e051d27ef4ceede (feat: add STT transcript echo toggle), included in v2026.7.7.

The duplicate discussion linking #13727 was useful context; that PR's maintainer follow-up also confirms that transcript echo is now on main. Thanks for the contribution.