feat(audio): add OpenRouter TTS and STT providers

[lfnovo]

Summary

Add OpenRouter as a first-class provider for TTS (text-to-speech) and STT (speech-to-text). OpenRouter is already first-class for LLM and embedding; it now exposes dedicated, OpenAI-compatible audio endpoints, so we can add the two remaining modalities.

Feasibility (verified)

OpenRouter ships OpenAI-compatible audio APIs:

• TTS → POST https://openrouter.ai/api/v1/audio/speech (compatible with OpenAI's Audio Speech API; returns a raw audio byte stream).
• STT → POST https://openrouter.ai/api/v1/audio/transcriptions (compatible with OpenAI's transcription API; returns JSON with text + usage).

Docs:

• TTS: https://openrouter.ai/docs/guides/overview/multimodal/tts
• STT: https://openrouter.ai/docs/guides/overview/multimodal/stt
• Announcement: https://openrouter.ai/blog/announcements/announcing-audio-apis/

Auth is the same OPENROUTER_API_KEY / https://openrouter.ai/api/v1 base URL already used by the existing OpenRouter LLM and embedding providers.

Implementation pointers

Because the endpoints are OpenAI-compatible, mirror the existing OpenAI audio providers rather than writing anything from scratch:

• TTS: add src/esperanto/providers/tts/openrouter.py following providers/tts/openai.py (or subclass/configure providers/tts/openai_compatible.py) with base URL https://openrouter.ai/api/v1 and OPENROUTER_API_KEY.
• STT: add src/esperanto/providers/stt/openrouter.py following providers/stt/openai.py (or providers/stt/openai_compatible.py) with the same base URL / key.
• Register in the factory (src/esperanto/factory.py): add "openrouter" entries under the stt and tts maps, alongside the existing openrouter LLM and embedding entries.
• Exports/__init__.py: export the new classes to match the pattern used by the other providers.

Acceptance criteria

• AIFactory.create_tts("openrouter", model_name=...) returns a working TTS model that synthesizes audio via OpenRouter.
• AIFactory.create_stt("openrouter", model_name=...) returns a working STT model that transcribes audio via OpenRouter.
• Both resolve credentials from OPENROUTER_API_KEY and the OpenRouter base URL, consistent with the existing OpenRouter LLM/embedding providers.
• Real-API integration coverage added in line with the existing STT/TTS release tests.

Downstream

Requested in Open Notebook: lfnovo/open-notebook#987. Once released here, Open Notebook only needs to add text_to_speech / speech_to_text to the OpenRouter modality mapping.