llm_dart_provider 0.11.0-alpha.1

Provider specification contracts and shared provider-facing data structures for llm_dart.

Changelog

Unreleased

• Added ProviderModelFacetSupport so provider facades can declare a narrower registry-visible model facet surface than their concrete class methods imply.
• Updated ProviderRegistry to respect provider-declared model facet support when listing providers and resolving dynamic model references.
• Added direct batching facts to EmbeddingModel and ImageModel contracts: maxEmbeddingsPerCall, supportsParallelCalls, and maxImagesPerCall.
• Expanded the shared image request contract with optional prompt support, aspect ratio, seed, file inputs, masks, and per-image provider metadata.
• Expanded SpeechGenerationRequest with shared outputFormat, instructions, speed, and language fields.
• Made provider-facing TranscriptionRequest.mediaType required so adapters no longer receive ambiguous audio bytes.
• Normalized ModelWarning with deprecated warning support plus stable feature and setting target fields while preserving legacy field serialization compatibility.

0.11.0-alpha.1 - 2026-05-12

• Alpha release of the provider contract package.
• Provides shared contracts for prompts, content parts, tools, results, metadata, model capabilities, and stream events.
• Includes UI message/projection types and JSON codecs for prompt, stream, and chat UI transport.
• Most app code reaches these types through llm_dart; custom provider authors can depend on this package directly.
• Treats model interfaces as provider implementation contracts: text models implement doGenerate(...) and doStream(...), embedding models implement doEmbed(...), and image, speech, and transcription models implement doGenerate(...).
• Defines the input/output boundary explicitly: provider-specific request controls belong in typed ProviderInvocationOptions or ProviderPromptPartOptions, while ProviderMetadata is reserved for output observations, replay data, and UI inspection.