Stabilize optimizer scoring and rubric consistency checks

[endymion]

Summary

• Prevent optimizer/test-created ScoreVersions from being marked featured by default.
• Add a score/rubric consistency preflight that compares a ScoreVersion's code against its own rubric, stores feedback-evaluation results, and exposes the command as plexus score contradictions.
• Improve optimizer baseline feedback evaluations so they request the consistency preflight.
• Improve evaluation/RCA dashboard filtering so category drill-downs can match score result IDs, item IDs, feedback item IDs, metadata IDs, and item identifiers.

Validation

• pytest plexus/score_rubric_consistency_test.py plexus/cli/score/scores_test.py plexus/cli/procedure/test_feedback_alignment_optimizer_config.py MCP/tools/evaluation/evaluations_test.py -q -> 71 passed
• flake8 plexus/score_rubric_consistency.py plexus/cli/score/scores.py plexus/cli/item/items.py plexus/cli/evaluation/evaluations.py MCP/tools/evaluation/evaluations.py -> passed
• cd dashboard && npm test -- --runTestsByPath components/__tests__/EvaluationTask.category-filter.test.tsx --runInBand -> 5 passed
• kbs validate -> passed
• Real smoke: plexus score contradictions --scorecard "SelectQuote HCS Medium-Risk" --score "Medication Review: Dosage" --version d730e6a8-164b-4157-b19b-47beb00d7b2e --format json returned potential_conflict.

Notes

• npm run typecheck initially failed on a corrupted generated .next/dev/types/routes.d.ts; after clearing .next, it ran for several minutes without completing and was stopped. CI should run the clean environment check.