0.10.2.md

0.10.2 release

Inference parameters are now backend-specific — each AI provider has its own typed params class where only explicitly set values are sent to the API, with an AdditionalParams dictionary for custom fields.