fix: clear duplicate issue discovery rows

[jakearmstrong59]

Summary

Fixes a duplicate-GitHub penalty leak where penalized evaluations could keep cached issue_discovery_issues rows and still consume issue-discovery repo allocation.

Problem

_zero_for_duplicate_penalty() cleared issue-discovery scalar fields, but did not clear MinerEvaluation.issue_discovery_issues.

After repo-level emission allocation, issue-discovery rewards are calculated from per-repo issue rows. A duplicate-penalized miner restored from cache could therefore have:

• issue_discovery_score = 0.0
• failed_reason set
• stale issue_discovery_issues still populated

Those stale rows could consume issue-discovery allocation before the later UID blacklist zeroing step.

Fix

• Clear issue_discovery_issues inside _zero_for_duplicate_penalty().
• Make emission allocation collectors ignore evaluations with failed_reason.
• Add regression coverage for duplicate-penalized cached issue rows.
• Add allocation tests proving failed PR and issue rows do not consume repo slices.

Tests

• .venv/bin/pytest tests/validator/test_duplicate_penalty_propagation.py tests/validator/test_blend_emission_pools.py -q
• .venv/bin/pytest tests/validator/test_github_identity_fallback.py -q
• .venv/bin/pytest tests/validator/oss_contributions/mirror/test_classes_integration.py -q
• .venv/bin/ruff check .
• .venv/bin/ruff format --check .
• git diff --check

Notes

Manual reproduction now shows duplicate-penalized miners have empty issue-discovery rows and receive no issue-discovery allocation.

Closes #1279