Feature: single daemon for hotkey dictation + OpenAI-compatible local STT API

[krystophny]

Problem

Running desktop hotkey dictation and API STT as separate daemons increases operational complexity and failure modes.

Proposal

Add a service mode to voxtype daemon that starts a local OpenAI-compatible transcription API in parallel with the existing daemon loop.

Goals

• Single daemon process for:
  • existing hotkey-driven dictation
  • local API transcription
• OpenAI-compatible endpoint:
  • POST /v1/audio/transcriptions
• Health endpoint:
  • GET /healthz
• Loopback-only by default (127.0.0.1)
• No built-in auth in v1 (assume trusted localhost)

Language behavior

• General language support remains configurable.
• Default constrained auto set for local deployment/tests: de,en.

Non-goals (v1)

• Public network exposure
• Built-in auth/mTLS
• TTS or chat endpoints

Why

This enables a clean architecture where upstream voxtype is the single local speech daemon, while another service (for example Tabura) handles external auth/proxy and forwards requests to localhost.

Acceptance criteria

• voxtype daemon can run hotkey flow and API flow concurrently.
• API transcriptions work while daemon is active.
• Existing daemon behavior remains unchanged when service mode is disabled.
• Integration tests cover multipart transcription endpoint and concurrent operation.