v2.25.0 — structured outputs + multi-modality profiles

Esperanto v2.25.0 — structured outputs everywhere, and OpenAI-compatible profiles that span every modality. Validated with the mocked canonical suite (1468 tests) and a clean-room build + install gate.

✨ New features

• Schema-driven structured outputs across all LLM providers. Set
  config={"structured": {"type": "json_schema", "schema": MyPydanticModel}} (or a
  JSON Schema dict) and read the parsed, validated result from response.structured.
  Works consistently across OpenAI, Azure, OpenAI-compatible, Google (Gemini), Vertex,
  Anthropic, Groq, Mistral, Ollama, OpenRouter, Perplexity, and Cohere. Each choice
  carries its own parsed value (choices[i].message.structured), so n>1 works.
  Schema mode is non-streaming in v1, and providers that can't honor a json_schema
  request fail fast with a clear error. (#95)
• Multi-modality OpenAI-compatible profiles. A registered
  OpenAICompatibleProfile can now serve embedding, speech_to_text, and
  text_to_speech in addition to language, via an opt-in capabilities field and
  per-modality default_models. All four factories resolve profiles through one shared
  precedence chain. Requesting an undeclared modality raises the new typed
  ProviderCapabilityError (unless a first-class class of the same name covers it —
  hybrid providers keep working). New base exception EsperantoError. (#230)
• requires_api_key flag for local/no-auth endpoints. A profile can now tolerate a
  missing API key (falls back to "not-required" instead of raising) — for local
  servers that need no auth. (#228)

🔌 New providers

• PayPerQ (PPQ) — a pay-as-you-go gateway across LLM + embedding + STT + TTS through
  one OpenAI-compatible endpoint. AIFactory.create_language("ppq", ...) (and the other
  three factories) with PPQ_API_KEY. (#238)
• Novita AI — pay-as-you-go gateway to many open-source models.
  AIFactory.create_language("novita", "moonshotai/kimi-k2.5") with NOVITA_API_KEY. (#120)
• OpenRouter text-to-speech and speech-to-text — rounds out OpenRouter to LLM + TTS +
  STT. (#224)
• oMLX — built-in profile for the local oMLX MLX
  server on Apple Silicon (language + embedding, no API key). (#228)

⚠️ Deprecations (back-compat kept)

1. OpenAICompatibleProfile(default_model=...) → use
   default_models={"language": ...}. The old field still works as a language alias and
   now emits a DeprecationWarning. No action required to keep current code working.

🙏 Thanks

• @JTCorrin — schema-driven structured outputs across all providers (#95)
• @kumar-shivang — OpenRouter TTS + STT providers (#224)
• @Alex-yang00 — Novita AI provider (#120)
• @mattius459 — PayPerQ provider (#238)
• @ukalia — pre-commit config with ruff + mypy hooks (#213)

And thanks to everyone who opened issues and drove the profile-mechanism design.

Full details in the CHANGELOG.
Install: pip install esperanto==2.25.0 (extras: esperanto[openai,anthropic,...]).