test(research): bind baseline vectors to production detector

[stacknil]

Summary

• add production-side exact conformance vectors for both episode research fixtures
• require the C++ detector to match committed baseline episode counts, identities, indexes, timestamps, and evidence IDs
• keep the change test-only and leave runtime, CLI, report, and fixture bytes unchanged

Why

The research baseline validator replays v0.6 selection and finding identity in Python. Without a production-side exact vector, a later C++ detector change could drift while the Python replay and static artifacts still remained internally consistent. These two vectors cover distinct boundaries: continuous low-density background with first-maximum selection, and a gap greater than the rule window that preserves two isolated episodes.

How to validate

• cmake --build build/ci-release --config Release -- /m:1
• ctest --test-dir build/ci-release -C Release --output-on-failure --timeout 120
• python -m ruff check scripts/episode_candidate_core.py scripts/episode_baseline_contract.py scripts/evaluate_episode_candidate.py tests/test_episode_candidate*.py
• python -B -m unittest discover -s tests -p "test_episode_candidate*.py" -v
• git diff --check origin/main...HEAD
• gitleaks git . --redact --no-banner --log-opts="origin/main..HEAD"

Risk / rollout notes

• Design decision: use exact production conformance vectors instead of a new test-only JSON driver; this is the smallest gate that detects detector algorithm or identity drift.
• Main risk: the raw C++ vectors intentionally duplicate the normalized fixture inputs, so a deliberate fixture change must update both sides.
• Compatibility impact: tests only; no source, runtime behavior, CLI, report schema, configuration, or release behavior changes.
• Rollback path: revert the single test commit; no migration or persisted state exists.

[stacknil]

Final current-head review for e110c22 is complete.

• Design decision: bind both normalized research baselines to exact production-side C++ detector vectors, avoiding a larger test-only JSON driver.
• Main risk: the raw C++ inputs intentionally duplicate the normalized fixture events; a deliberate fixture revision must update both conformance surfaces.
• Compatibility impact: test-only. Runtime source, CLI behavior, loglens.report.v3, configuration, fixture bytes, and released v0.6 behavior are unchanged.
• Rollback path: revert the single test commit; no migration or persisted state exists.

Exact-head evidence:

• GitHub checks: 11/11 successful; merge state CLEAN
• MSVC warnings-as-errors build: passed
• CTest: 7/7 passed
• Episode-candidate research suite: 25/25 passed
• Ruff: passed
• git diff --check: passed
• Gitleaks: no leaks
• Final one-file / 79-line diff review: no unresolved reuse, quality, efficiency, clarity, privacy, or scope findings