feat: add ThinkingChunk to StreamChunk protocol

[tony-chinchill-ai]

Summary

• Add ThinkingChunk dataclass (type="thinking", content="") to StreamChunk union for AI reasoning display
• Slack adapter: extract content field in send_structured_chunk() so thinking is serialized to Slack's streaming API
• Teams adapter: render thinking chunks as italic text above accumulated response
• Additive-only — consumers that don't handle "thinking" chunks silently ignore them

Context

Bridges pydantic-ai thinking events (Chinchill-AI/chinchill-api#952) to SDK StreamChunks so agent reasoning chains (e.g. "Analyzing → Checking → Resolving") are displayed rather than just logged.

Test plan

• 3 new tests for ThinkingChunk dataclass (defaults, content, literal type)
• 2 new tests for Teams stream thinking chunk handling (italic rendering, skip-when-no-text)
• Full suite passes (3307 passed, 0 failures)
• Lint clean on changed files

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request introduces the ThinkingChunk type to support the display of AI reasoning and thinking processes across different chat platforms. The Slack and Teams adapters have been updated to handle these new chunks, with the Teams adapter specifically rendering thinking content as italicized text above the main message. The changes also include comprehensive unit and integration tests. A review comment suggests refactoring the chunk processing logic in the Teams adapter to improve readability and reduce redundant attribute checks.

[gemini-code-assist]

This series of elif statements can be refactored to reduce redundancy and improve readability. Specifically, the checks for markdown_text are duplicated, and the hasattr checks for different chunk types can be grouped together.

Suggested change

	elif isinstance(chunk, dict) and chunk.get("type") == "markdown_text":
	text = chunk.get("text", "")
	elif hasattr(chunk, "type") and getattr(chunk, "type", None) == "markdown_text":
	text = getattr(chunk, "text", "")
	elif hasattr(chunk, "type") and getattr(chunk, "type", None) == "thinking":
	thinking = getattr(chunk, "content", "")
	# Render thinking as italic context above the main text
	if accumulated or not message_id:
	continue # Will be included in next text update
	elif isinstance(chunk, dict) and chunk.get("type") == "markdown_text":
	text = chunk.get("text", "")
	elif hasattr(chunk, "type"):
	chunk_type = getattr(chunk, "type", None)
	if chunk_type == "markdown_text":
	text = getattr(chunk, "text", "")
	elif chunk_type == "thinking":
	thinking = getattr(chunk, "content", "")
	# Render thinking as italic context above the main text
	if accumulated or not message_id:
	continue # Will be included in next text update

[tony-chinchill-ai]

Closing: ThinkingChunk (and related TaskUpdateChunk, PlanUpdateChunk) don't exist in the upstream Vercel chat-sdk. These will be removed from the library and any custom chunk types will live in application code instead.