Add repeated trials and stability scoring

[askmy-stack]

Summary

Support repeated model-backed probe trials and summarize result stability.

Acceptance criteria

• Trial count is configurable and defaults conservatively
• Results retain per-trial details and aggregate summary
• Stability score accounts for tool selection and argument validity
• Reports highlight unstable probes separately from deterministic failures
• Seed/configuration metadata is recorded when the provider supports it
• Tests cover stable, unstable, and partial-failure runs