Transcribe speech from a managed audio file using faster-whisper. Use this when the user asks to extract text from audio.

Inputs:
- path: required path relative to tools.file_storage.root_dir
- language: optional language hint (for example "en")
- task: optional "transcribe" or "translate"

Returns transcript text, language metadata, duration, and timestamped segments.

If the user has uploaded an audio file, use its managed path from incoming files metadata.
