daily-fix: issue658 test sys.modules stub leak

Overview / Motivation

Auto-filed by the /daily 2026-07-12 Step C parked-candidate routing pass, from the prose follow-up parked on task #1037 (2026-07-05T13:04:24Z; emitting agent: #1037's implementer). NON-workflow-surface (experiment test hygiene) — filed daily-auto-filed only, no wf-fix tags.

Goal

Fix the sys.modules-stub isolation bug in tests/test_issue658_judge_e0_batch_migration.py — the stub leaks across test files in the same pytest session.

Bug

• Observed: _install_issue658_common_stub() registers a fake issue658_common module in sys.modules at import time (MOD = _load_module() at module scope) and never removes it; the guard if "issue658_common" in sys.modules: return also means a REAL issue658_common loaded earlier silently replaces the stub (and vice versa: any later test importing issue658_common in the same session gets this file's trivial stand-ins).
• Confidence: medium.
• verified-at-filing: sed -n '50,80p' tests/test_issue658_judge_e0_batch_migration.py (2026-07-13) — stub installed at module scope, no fixture teardown, early-return guard present.

Proposed change (refine in planning)

Scope the stub with a proper fixture (install in a module-scoped fixture with teardown restoring the prior sys.modules entry), or namespace the module under test so the stub never occupies the shared issue658_common key.

Scope / surfaces

• tests/test_issue658_judge_e0_batch_migration.py (experiment test — ordinary infra code-change pipeline applies).

Constraints / invariants

• Tests keep passing in isolation AND in full-suite order; no ordering-dependent behavior remains.

Provenance

Origin: prose follow-up parked on #1037 events.jsonl at 2026-07-05T13:04:24Z ("tests/test_issue658_judge_e0_batch_migration.py sys.modules-stub isolation bug (stubs leak across test files in the same pytest session)"), re-verified at filing time.