Pin recovery flow to a primary execution target

[mmprotest]

Motivation

• Prevent the runner from silently pivoting to a different execution/validation target after a hard syntax/import/runtime failure and keep recovery focused on the original failing target.
• Introduce a minimal, generic guard scoped to recovery behavior without adding task-specific heuristics or redesigning the runner.

Description

• Added a minimal runtime field _primary_execution_target to Runner state and initialized/reset it alongside existing recovery/active-solution fields in villani_code/state.py.
• Introduced helpers _primary_execution_target and _seed_primary_execution_target and seeded the primary target the first time a meaningful validation/launch/active-solution target is discovered in villani_code/state_runtime.py.
• During recovery, blocked Bash validation/launch attempts that would switch to a different target by comparing attempted target vs primary_execution_target and returning compact structured feedback from the existing small_model_tool_guard path in villani_code/state_runtime.py.
• On hard failure, pinned/seeded the primary execution target, included it in the emitted recovery activation event payload, and ensured _recovery_mode remains focused on that primary target.
• Made recovery clearing conservative: PASS/REPAIRED only clears recovery when the primary target itself was validated (or if no primary target exists), and updated completion gating to prefer primary_execution_target when checking validation completeness.
• Tests: added and updated unit tests that cover primary seeding, hard-failure pinning, blocked switch attempts, allowed helper edits, allowed rerun of same primary, and conservative recovery clearing.

Testing

• Ran: pytest -q tests/test_state_runtime.py tests/test_state_tooling_mutation_policy.py tests/test_bounded_execution.py and observed all tests passing.
• Outcome: 64 passed for the executed test subset.

---

Codex Task