v2.25.1 — auth & interop fixes

A maintenance patch: three bug fixes to authentication, LangChain interop, and provider defaults. Validated with the canonical mocked suite (1480 passed), lint, types, and a clean-room build + import gate.

🐛 Notable fixes

• Vertex AI authenticates from a service-account key across all modalities (#249). VertexTextToSpeechModel and VertexEmbeddingModel previously ignored a configured key and shelled out to the gcloud CLI — unusable in containers where gcloud isn't installed. All three Vertex providers (LLM, TTS, embedding) now share one auth path (VertexAuthMixin): explicit key file (credentials_file / credentials_path config, or GOOGLE_APPLICATION_CREDENTIALS) → Application Default Credentials → gcloud as a last resort. A missing gcloud binary now raises a clear error instead of an opaque crash.
• to_langchain() forwards a custom base URL for Anthropic and Cohere (#229). Converting an Anthropic- or Cohere-compatible model configured with a custom base_url to LangChain previously reconnected silently to the official API. The official endpoint is still omitted so LangChain keeps its own default.

⚠️ Behavior change

• ElevenLabs default models bumped to the latest (#245). Speech-to-text now defaults to scribe_v2 (was scribe_v1); text-to-speech now defaults to eleven_v3 (was eleven_multilingual_v2). Both older models remain usable by passing an explicit model_name. Verified against the live ElevenLabs API. If you relied on the previous defaults implicitly, pin them with model_name.

🙏 Thanks

All fixes this cycle were reported and shipped in-house. Thanks to everyone opening issues and testing against real provider endpoints — that feedback is what keeps the parity promise honest.

Full details in the CHANGELOG.
Install: pip install esperanto==2.25.1 (extras: esperanto[openai,anthropic,...])