Audio transcription lacks diagnostic logging

Summary

Audio transcription via Deepgram (or other providers) happens silently without any diagnostic logging, making it impossible to debug transcription failures.

Environment

• OpenClaw version: 2026.1.29 (dev channel)
• Audio config: tools.media.audio.enabled: true, provider: deepgram, model: nova-3
• Diagnostics flags include: audio.*

Problem

When a voice message arrives (e.g., from Telegram), the transcription either works or silently fails. There are no logs indicating:

1. Whether a file download was attempted
2. Whether Deepgram API was called
3. Success/failure of transcription
4. Any errors or reasons for skipping transcription

Reproduction

1. Send two voice messages to the bot in quick succession
2. First message: no transcript appeared in agent context
3. Second message: transcript appeared correctly ([Audio] Transcript: ...)
4. Logs show no difference between the two — no audio-related entries at all

Expected Behavior

With audio.* in diagnostics flags, expect to see logs like:

[audio] downloading file from telegram: file_id=xxx
[audio] deepgram transcription start: provider=deepgram model=nova-3
[audio] deepgram transcription complete: duration=1.2s chars=45

Or on failure:

[audio] deepgram transcription failed: error="rate limit" status=429
[audio] skipping transcription: reason="file download failed"

Impact

Without logging, users cannot:

• Debug why some voice messages aren't transcribed
• Detect Deepgram API issues (rate limits, auth failures)
• Verify audio pipeline is working correctly

Suggested Fix

Add subsystem logging to the audio transcription pipeline, gated by audio.* diagnostic flag.