Add paired distance venue-regime evidence

[khtad]

Summary

• Add paired/stratified distance-location evidence in src/venue_bias.py, comparing visiting-team corrected distances at a venue against those same teams' away shots elsewhere in the same season.
• Wire the DB-backed venue scorecard runner to build the diagnostic from in-memory prior-corrected distances without mutating shot_events or venue_bias_corrections.
• Add a Distance-Location Paired Diagnostics section to the Markdown artifact and refresh the live artifact.
• Refresh roadmap/component docs plus the venue-scorekeeper-bias KB page, index, and log.

Residual-regime failure

The residual gate still fails in regime-aware mode, but not because max |z| >= 2 is an automatic veto. The remaining blockers are candidate venue-seasons classified as blocking (unexplained_or_confounded, population_shift_detected, or insufficient_evidence) rather than supported persistent_bias or temporary_supported_regime.

Latest artifact result:

• Held-out log-loss: PASS (delta = -0.000015)
• Home-ice over-correction: PASS (removed = -0.013, limit 0.500)
• Distance/location residuals: FAIL (10 blocking regimes, 18 supported regimes, max |z| = 4.067)
• Event-frequency residuals: FAIL (5 blocking regimes, 22 supported regimes, max |z| = 3.572)
• Distance paired diagnostics: 28 candidate residuals, 17 supported by paired visiting-team evidence

Verification

• wsl bash -lc "python3 -m venv /tmp/test-venv && /tmp/test-venv/bin/pip install -q pytest requests numpy scipy scikit-learn && /tmp/test-venv/bin/python -m pytest -q tests/test_venue_bias.py tests/test_validation.py tests/test_venue_correction_validation_export.py tests/test_venue_correction_validation_from_db.py" -> 81 passed
• python scripts/check_knowledge_base_update.py --base-ref origin/main -> passed
• git diff --check origin/main...HEAD -> clean
• git diff --check -> clean

Knowledge base

Updated knowledge_base/wiki/concepts/venue-scorekeeper-bias.md, knowledge_base/index.md, and knowledge_base/log.md for the paired/stratified distance evidence and refreshed live scorecard result.