fix: use cached issue-discovery scores when DAS mirror fetch fails (#1065)

[alpurkan17]

Summary

When a validator cannot fetch issue-discovery data for one miner from the DAS GitHub mirror, that miner is skipped for the round and receives a zero issue-discovery score. The remaining miners are then normalized across the full pool, unfairly redistributing the failed miner's share.

This fix adds a mirror_issue_fetch_failed flag on MinerEvaluation, set by mirror_scan.py when MirrorRequestError is raised. During issue_discovery() in forward.py, the validator restores the previous round's cached issue-discovery scores for that miner *before* normalization runs, so a transient DAS fetch failure does not distort emission distribution.

The MinerEvaluationCache.store() method also preserves previous issue-discovery scores when the current eval has zero (since the cache is stored before issue discovery runs in step 1 of forward()). A new update_issue_discovery_scores() method refreshes the cache for successful miners after discovery completes.

Validation

• mirror_issue_fetch_failed is set on MinerEvaluation when MirrorRequestError is caught in mirror_scan.py
• Cached scores are restored before normalize_issue_discovery_rewards() runs — emission distribution is correct
• Cache preserves previous issue-discovery scores in store() as a safety net
• update_issue_discovery_scores() refreshes cache entries for successful miners
• ruff check + ruff format pass

Closes #1065

[alpurkan17]

Competitor #1066 already merged. Closing.