feat: Phase X X1 multi-batch learning record schema (MBL-)

[cschanhniem]

Summary

• Adds MBL- (multi-batch learning record) schema: per-batch snapshot of prediction quality after wet-lab feedback
• Auto-computes hit_rate from n_confirmed_hits/n_candidates_tested; quality_grade from hit_rate (A≥40%, B≥25%, C≥10%, D<10%, N/A when no wet-lab data)
• AUROC field for ranking quality; whr_ids list linking to wet-lab hit records; pcu_id for calibration update link
• no_wet_lab_data status forces N/A grade (cannot claim quality without data)
• Adds Phase X (X1-X5) definition to NEXT_100_PR_MAP: multi-batch learning loop phase
• dry_lab_only=True enforced; 61 tests

Test plan

• python3 -m pytest tests/evidence/test_multi_batch_learning_record.py -- 61 passed