fix: alert on mirror service outage instead of silently zeroing scores

[wdeveloper16]

Summary

Follow-up to #796. When mirror.gittensor.io is unavailable, fetch failures were either logged at WARNING level (issue discovery) or without scoring-impact context (OSS path), making it impossible to distinguish a service outage from isolated miner failures by watching error logs.

• Upgrade per-miner issue-discovery fetch failure from WARNING → ERROR (consistent with OSS path severity)
• Improve OSS per-miner error message to include hotkey prefix and explicit consequence ("scores will be 0 this round for this miner")
• Add aggregate outage detection in run_mirror_issue_discovery: emits ERROR when ≥50% of eligible miners fail and total_eligible >= 5 (the guard suppresses false positives in small/dev environments)
• Add 4 tests covering: total failure fires alert, 50% boundary fires, 20% rate does not fire, small-population guard suppresses

No behaviour changes

Scoring logic is untouched — this is purely observability. All 51 existing tests continue to pass.

Test plan

• uv run --with pytest pytest tests/validator/oss_contributions/mirror/test_load.py tests/validator/issue_discovery/test_mirror_scan.py -v → 51 passed

[anderdc]

Needless logging change. Closing