Add reasoning (extended thinking) for claude 3.7

[statefb]

Issue #, if available:
close #745

Description of changes:

This PR adds support for Claude 3.7's reasoning feature (extended thinking), allowing users to see the model's step-by-step thought process when generating responses.

Backend Changes

Core Functionality

• Added new ReasoningContentModel to represent reasoning content returned by Claude
• Extended the streaming handler to process reasoning blocks from the API response
• Added enable_reasoning parameter to conversation APIs
• Fixed type definitions to handle the new content types
• Implemented proper state management for reasoning (similar to agent thinking)
  • Pass all reasoning content (text, signature, redacted) in every streaming. See detail:
    • https://docs.anthropic.com/en/docs/about-claude/models/extended-thinking-models

Streaming Handler Updates

• Added PartialReasoningContent type definition to handle reasoning blocks in the stream
• Added on_reasoning callback to dispatch reasoning tokens to frontend
• Stream handler now processes both regular responses and reasoning blocks separately

Bot Configuration

• Added reasoning_params to GenerationParams with budget_tokens setting
• Set reasonable defaults for reasoning budget (1024 tokens, which is minimum value)
• Updated bot repository code to handle and persist the reasoning parameters
• Made necessary adjustments to ensure backward compatibility with existing bots

API & Input

• Extended ChatInput schema with enable_reasoning flag

Testing

• Added new tests for reasoning functionality
• Added test environment variables to invoke us-west-2 model

Disclaimers

• Used # type: ignore for ReasoningModel due to incomplete boto3-stubs
• Updated boto3 to version 1.37.0 to support the reasoning feature

Frontend Changes

UI Components

• Added ButtonReasoning component to toggle reasoning display
• Created ReasoningCard component to display reasoning content
• Updated ChatMessage component to handle reasoning content
• Enhanced input component with reasoning toggle button

State Management

• Added reasoning state management using XState
• Integrated reasoning into useChat hook
• Added model-specific reasoning support detection

Configuration

• Added budget tokens configuration in the generation parameters UI
• Updated token limits to accommodate Claude 3.7's capabilities
• Added relevant translations for reasoning UI elements

Disclaimers

• Used XState for reasoning which may be considered heavyweight for this use case, but it aligns our existing agent implementation and allows for future expansion
• Added reasoning state to useChat hook to leverage existing message posting functionality and avoid duplication

Other Changes

• Added Pyright configuration to exclude specific directories
• Updated model defaults to support Claude 3.7 features
• Added proper type definitions throughout the codebase

Streaming response example (convserse API)

event: {'messageStart': {'role': 'assistant'}}
event: {'contentBlockDelta': {'delta': {'reasoningContent': {'text': 'この'}}, 'contentBlockIndex': 0}}
event: {'contentBlockDelta': {'delta': {'reasoningContent': {'text': 'パズルのような'}}, 'contentBlockIndex': 0}}
event: {'contentBlockDelta': {'delta': {'reasoningContent': {'text': '問いかけを'}}, 'contentBlockIndex': 0}}
event: {'contentBlockDelta': {'delta': {'reasoningContent': {'text': '解いていきましょう。'}}, 'contentBlockIndex': 0}}
...
event: {'contentBlockDelta': {'delta': {'reasoningContent': {'signature': 'EqgBCkYQARgCIkD...'}}, 'contentBlockIndex': 0}}
event: {'contentBlockStop': {'contentBlockIndex': 0}}
event: {'contentBlockDelta': {'delta': {'reasoningContent': {'redactedContent': b'ErQBCkYQARgCKkA...'}}, 'contentBlockIndex': 1}}
event: {'contentBlockStop': {'contentBlockIndex': 1}}
event: {'contentBlockDelta': {'delta': {'text': 'このなぞな'}}, 'contentBlockIndex': 5}}
event: {'contentBlockDelta': {'delta': {'text': 'ぞの答えは、'}}, 'contentBlockIndex': 5}}
...

Screenshots

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.