Close [AI-86](https://linear.app/affine-design/issue/AI-86)
<!-- This is an auto-generated comment: release notes by coderabbit.ai -->
## Summary by CodeRabbit
- **New Features**
- Improved chat session forking to allow creating a fork without specifying the latest message, enabling more flexible session management.
- **Bug Fixes**
- Forking a chat session with an invalid latest message ID now correctly returns an error.
- **Tests**
- Added and updated test cases to cover session forking with missing or invalid latest message IDs, ensuring robust behavior in these scenarios.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
Support [AI-82](https://linear.app/affine-design/issue/AI-82).
Added a `TextStreamParser` class to standardize formatting of different types of AI stream chunks across providers.
### What changed?
- Created a new `TextStreamParser` class in `utils.ts` that handles formatting of various chunk types (text-delta, reasoning, tool-call, tool-result, error)
- Refactored the Anthropic, Gemini, and OpenAI providers to use this shared parser instead of duplicating formatting logic
- Added comprehensive tests for the new `TextStreamParser` class, including tests for individual chunk types and sequences of chunks
- Defined a common `AITools` type to standardize tool interfaces across providers
<!-- This is an auto-generated comment: release notes by coderabbit.ai -->
## Summary by CodeRabbit
- **New Features**
- Enhanced formatting and structure for streamed AI responses, including improved handling of callouts, web search, and web crawl results.
- **Refactor**
- Streamlined and unified the processing of streamed AI response chunks across providers for more consistent output.
- **Bug Fixes**
- Improved error handling and display for streamed responses.
- **Tests**
- Added comprehensive tests to ensure correct formatting and handling of various streamed message types.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
fix AI-4
fix AI-18
better provider/model choose to allow fallback to similar models (e.g., self-hosted) when the provider is not fully configured
split functions of different output types
fix AI-20
<!-- This is an auto-generated comment: release notes by coderabbit.ai -->
## Summary by CodeRabbit
- **New Features**
- Enhanced file metadata with MIME type, blob ID, and file name across context and workspace, now visible in UI and API.
- Added workspace-level matching for files and documents with configurable thresholds and workspace scoping in search queries.
- Introduced a new error type and user-friendly messaging for global workspace context matching failures.
- **Bug Fixes**
- Improved consistent handling of file MIME types and nullable context IDs for accurate metadata.
- **Documentation**
- Updated GraphQL schema, queries, and mutations to include new metadata fields, optional parameters, and error types.
- **Style**
- Added new localization strings for global context matching error messages.
- **Tests**
- Extended test coverage with new and updated snapshot tests for metadata and matching logic.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
Close [AI-59](https://linear.app/affine-design/issue/AI-59)
<!-- This is an auto-generated comment: release notes by coderabbit.ai -->
## Summary by CodeRabbit
- **New Features**
- Added support for two new providers, Anthropic and Exa, in test environments and configurations.
- **Chores**
- Updated workflow and test environment setup to include API keys for Anthropic and Exa providers.
- **Refactor**
- Changed the AI model used for the "Chat With AFFiNE AI" prompt to "claude-3-7-sonnet-20250219".
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
<!-- This is an auto-generated comment: release notes by coderabbit.ai -->
## Summary by CodeRabbit
- **New Features**
- Expanded file chunk matching to include both context and workspace file embeddings, providing broader and more relevant search results.
- **Improvements**
- Enhanced result ranking by introducing a re-ranking step for combined embedding matches, improving the relevance of returned file chunks.
- Adjusted file count reporting to reflect the total number of workspace files instead of ignored documents for more accurate workspace file statistics.
- Renamed and streamlined workspace file management methods for clearer and more consistent API usage.
- **Bug Fixes**
- Prevented embedding similarity queries when embedding is disabled for a workspace, improving system behavior consistency.
- **Tests**
- Added comprehensive tests to verify workspace embedding management, including enabling, matching, and disabling embedding functionality.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
DarkSky
