* Add typed MCPCallToolResponse interface and format converters
The commit introduces a structured response type for MCP tool calls with
proper handling for different content types (text, image, audio). It adds
provider-specific converter functions to transform tool responses into
the appropriate message format for OpenAI, Anthropic, and Gemini.
* Support vision models in tool call responses
Add isVisionModel parameter to tool response formatters to conditionally
handle content based on model capabilities. Non-vision models now receive
JSON stringified content, while vision models get proper multimedia parts.
LiuVaayne
a6374abcb9