fix: validate screened candidate IDs without totals

[Stahl-G]

Summary

• Fix the fix: close topology review gaps #350 post-merge review gap where screened candidate ID validation only ran when total_candidates / candidate count metadata was declared.
• When candidate_claims.json has stable IDs, selected/excluded/deprioritized screened entries must now carry known, non-duplicate candidate_id values even if the screening policy only has fields such as max_items.
• Keep full universe coverage checks tied to declared totals, so this does not broaden the optional total/count contract.

Boundary

This is a validator hardening PR only. It does not change topology, stage order, runtime handoff guidance, Claim Ledger behavior, gates, delivery, or release authority.

Validation

• python3 -m pytest -q tests/test_runtime_state.py -k 'screened_candidates or candidate_universe' -> 22 passed
• python3 -m pytest -q tests/test_runtime_state.py -> 300 passed
• python3 scripts/check_release_consistency.py --no-tag
• git diff --check