0.21

• New model: o3-mini. #728
• The o3-mini and o1 models now support a reasoning_effort option which can be set to low, medium or high.
• llm prompt and llm logs now have a --xl/--extract-last option for extracting the last fenced code block in the response - a complement to the existing --x/--extract option. #717