test(research): lock maximal-window tie break

[stacknil]

Summary

• add a minimal equal-score candidate control for the final chronological tie-break
• verify candidate input order does not change the selected window
• record the bounded evidence and remaining research boundaries in the episode ADR

Why

The v0.7 research matrix named maximal-window ties as an unverified policy point. Six synthetic events are sufficient to create exactly two candidates with equal covered-event count, total span, and episode count, so the final chronological key is the only distinguishing objective. A focused control gives higher decision value than another large JSON fixture.

How to validate

python -m unittest tests.test_episode_candidate_core
python -m unittest discover -s tests -p 'test_episode*.py'
python -m ruff check scripts tests
cmake --build build --config Debug --target test_detector
ctest --test-dir build -C Debug --output-on-failure

Risk / rollout notes

• Design decision: when all higher-priority objectives tie, select the lexicographically earliest chronological window key; candidate input order must not affect the result.
• Main risk: overgeneralizing one bounded characterization case to unresolved shared-evidence or production-complexity policy.
• Compatibility impact: test and ADR only; Detector::analyze(), CLI behavior, loglens.report.v3, and v0.6 behavior are unchanged.
• Rollback path: revert commits 2008d8e and 4d01850; no migration or runtime rollback is required.

[stacknil]

Post-CI final diff review completed.

• Checks: all 11 reported status entries succeeded (Ubuntu, Windows, CodeQL, parser fuzz smoke, and Repo Sentinel; push and pull-request workflows both ran).
• Design decision: when covered-event count, total span, and episode count tie, the lexicographically earliest chronological window key wins; reversing candidate input must not change the selection.
• Main risk: treating this bounded characterization as evidence for unresolved shared-evidence or production-complexity policy. The ADR keeps those boundaries explicit.
• Compatibility impact: the final diff contains only tests/test_episode_candidate_core.py and docs/adr/0001-episode-semantics-boundaries.md; there is no src/, CLI, report-v3, rule, or v0.6 behavior change.
• Rollback path: revert 2008d8e and 4d01850; no runtime migration or data rollback is required.

No review comments, bot action items, privacy findings, or secret findings remain.