fix: harden public correctness before v0.1.2

[mireya-o]

Summary

• align the public runtime contract to Python 3.11+
• fix deterministic grading for tool_error events
• reject invalid diff comparisons when suite identity or case sets do not match
• add targeted tests for the corrected blocking paths

Why

The public repo was still declaring Python 3.10+ while suite parsing requires stdlib tomllib in practice, and the blocking path needed stricter correctness before any real external validation.

Validation

• ruff check .
• pytest -q
• fresh-clone README quickstart:
  • run exit = 0
  • diff exit = 0
• fresh-clone golden path demo:
  • status = SUCCESS
  • expected pass/fail diff behavior observed