Report tool-selection and argument-validity metrics

[askmy-stack]

Summary

Add reportable metrics for model-backed behavioral runs so teams can compare baseline and candidate tool interfaces.

Acceptance criteria

• Report tool-selection accuracy by probe and aggregate
• Report argument-validity rate and validation failures
• Report risk and confirmation expectation compliance
• Results are available in JSON and Markdown
• Metrics distinguish missing data from failed evaluations
• Tests use deterministic fixture results