feat: add voice message transcription support

[kky]

Summary

• Telegram voice messages are now downloaded and transcribed when the /add-transcription skill is installed
• Uses dynamic import('../transcription.js') — no compile-time dependency on the transcription module
• Falls back gracefully with a helpful message when transcription is not available
• Agent responses to voice messages begin with "You said: "

How it works

• The message:voice handler downloads the audio file via Telegram Bot API (ctx.getFile() + file URL)
• Dynamically imports transcribe() from src/transcription.ts (installed via /add-transcription)
• If the module is missing (MODULE_NOT_FOUND), delivers a placeholder message suggesting the user run /add-transcription
• If transcription fails at runtime, delivers a failure message
• Uses @ts-ignore on the dynamic import since the module is optional

Tests

• Test for graceful fallback when transcription skill is not installed
• Test for successful transcription when skill is available (skips automatically when module not present)
• All existing telegram tests continue to pass

How it was tested

• Multiple clean install tests: fresh clone → add telegram → verify text works → add transcription → verify voice transcription works
• Tested with both local whisper.cpp and OpenAI Whisper API backends
• Tested on Linux (Ubuntu, aarch64)