v4.8
API: Move OpenAI-compatible API from extensions/openai to modules/api
API: Add thinking block signature field, fix error codes, clean up logging
API: Fix crash on list-format content in tool and assistant messages
Remove hardcoded trust_remote_code=True in embedding loader
Image generation: Embed generation metadata in API image responses
API: add instruction_template support to the model load endpoint
API: Fix IPv6 address formatting
Add `preserve_thinking` chat template parameter (UI and CLI)
API: Fix _start_cloudflared raising after first attempt instead of exhausting retries