v0.2.19

Fixed

• BAML backend now preserves provider-specific usage fields (e.g., cache_creation_input_tokens, cache_read_input_tokens) returned by the LLM provider alongside the default input_tokens and output_tokens