audio: Telegram voice message and audio file handling

[bug-ops]

Parent: #520
Depends on: #521, #522

Context

Telegram supports voice messages (OGG/Opus) and audio file attachments natively. teloxide already provides `Message::voice()` and `Message::audio()` accessors — currently ignored.

Changes

`crates/zeph-channels/src/telegram.rs`

• Check `msg.voice()` and `msg.audio()` in addition to `msg.text()`
• Download audio via Telegram Bot API `getFile` + HTTP fetch
• Convert to `Attachment::Audio` and attach to `ChannelMessage`
• If text is also present (caption), include both

Audio flow

1. User sends voice message in Telegram
2. Adapter detects `msg.voice()`, gets `file_id`
3. Downloads OGG/Opus file via Bot API
4. Creates `ChannelMessage { text: "", attachments: [Audio { data, mime: "audio/ogg" }] }`
5. Agent loop transcribes via configured STT backend
6. Transcribed text enters normal agent pipeline

Size limits

• Telegram voice messages: up to 20 MB
• Apply configurable max duration (default: 120s) to prevent abuse

Acceptance criteria

• Voice messages transcribed and processed as text commands
• Audio file attachments (.mp3, .ogg, .wav) supported
• Caption text combined with transcription
• File size/duration limits enforced
• Graceful error message if STT is not configured

[bug-ops]

Implemented in PR #562 (feat/audio-input-2)