Gate completion on active-solution validation and restrict full rewrites during recovery

[mmprotest]

Motivation

• The runner could mark a mission as completed even when the active solution file still failed validation, and it could perform broad full-file rewrites of that active file during recovery; this change makes completion gating and edit discipline conservative and surgical.
• Keep behavior minimal and generic: track the most recent validation of the active solution file, block mission success on failed/unknown validation, and prevent one-shot full rewrites of that file during recovery while allowing bounded patches and helper-file edits.

Description

• Added active-solution validation state to the runtime (_active_solution_last_validation_ok, _active_solution_last_validation_summary) and initialize/reset them with run state (in Runner.__init__ and run reset). (file: villani_code/state.py)
• Gate final completion on that state by checking it in the run-time completion guard (_budget_reason) and funneling final textual/no-tool completion branches through _finish_bounded so the termination reason reflects blocked completion when validation is missing/failed. (file: villani_code/state.py, function: Runner.run internals / _budget_reason / completion paths)
• Enhance verification flow to detect whether a validation/run command targeted the active solution file and mark _active_solution_last_validation_ok = False whenever the command shows a hard failure signal (non-zero exit or crash tokens like SyntaxError, NameError, ImportError, ModuleNotFoundError, traceback/unhandled exception); set it back to True on clean validation. Added helper utilities _command_targets_active_solution and _has_hard_failure_signal. (file: villani_code/state_runtime.py, function: run_verification + helpers)
• Tighten recovery edit guard: when _recovery_mode is active and the active solution file has a recent hard failure (_active_solution_last_validation_ok is False), block broad full-file Write replacements of the active file and return compact structured JSON feedback; allow bounded Patch edits and edits to helper files. (file: villani_code/state_runtime.py, function: small_model_tool_guard)
• Tests updated/added to validate behavior and discipline: assertions added to verification tests for active-solution validation state, added bounded execution tests to ensure completion is suppressed when validation is failing and allowed when validation is clean, and added mutation-policy tests to assert full-write blocking and bounded-patch allowance. (files: tests/test_state_runtime.py, tests/test_bounded_execution.py, tests/test_state_tooling_mutation_policy.py)

Testing

• Ran the targeted test set: pytest -q tests/test_state_runtime.py tests/test_state_tooling_mutation_policy.py tests/test_bounded_execution.py and all tests passed (61 passed).
• The modified tests exercise: failing validation detection (SyntaxError/NameError/ImportError), recovery-mode state transitions, completion gating when active-solution validation is false/true, and recovery write/patch blocking behavior; they all passed.

---

Codex Task