fix: don't cache fabricated zero when mirror scoring_data_stored=False

[ai-mountain]

Summary

In _resolve_solving_pr_score (gittensor/validator/issue_discovery/mirror_scan.py), a mirror response with scoring_data_stored=False was treated as a successful fetch of an empty diff — mirror_files_to_legacy() returned no file changes, calculate_base_score_for_pr_files() returned base_score=0 / token_score=0, and that fabricated zero was committed to the cross-miner solving-PR cache. Sibling miners' lookups against the same (repo, pr_number) then served the cached zero with no in-cycle retry, even after the mirror finished its backfill mid-cycle.

This PR adds an explicit guard immediately after the MirrorRequestError branch: when files_response.scoring_data_stored is False, increment cache_stats.fetch_failures, log a warning that mentions data unavailability (not "below threshold"), and return None without writing to cache. Parity with the existing MirrorRequestError handling — same three actions, same callsite-visible semantics.

Scope: only gittensor/validator/issue_discovery/mirror_scan.py and its sibling test file. No change to mirror/scoring.py and no change to credibility/eligibility behavior

Related Issues

Closes #836

Type of Change

• Bug fix
• New feature
• Refactor
• Documentation
• Other (describe below)

Testing

• tests/validator/issue_discovery/test_mirror_scan.py::TestCacheStats — 2 new cases:
  • test_resolve_treats_unavailable_scoring_data_as_failure — single call: asserts result is None, fetch_failures == 1, empty cache.
  • test_unavailable_scoring_data_is_not_cached_across_sibling_lookups — the literal acceptance criterion from [Bug] Mirror issue-discovery caches a fabricated zero score when scoring_data_stored=False #836: two issues sharing one solving PR, scoring_data_stored=False returned both times → misses == 2, fetch_failures == 2, cache == {}, client.get_pr_files.call_count == 2.
• Full validator test suite passes locally (728/728).
• Ruff lint + format clean (uv run pre-commit run --all-files).
• Pyright: 0 errors / 0 warnings (uv run pre-commit run --all-files --hook-stage pre-push).

Checklist

• Code follows project style guidelines.
• Self-review completed.
• No change to mirror OSS scoring (oss_contributions/mirror/scoring.py).
• No change to credibility/eligibility behavior; legacy parity preserved.
• Cache mutation is gated: cache stays empty on availability failure, leaving the slot open for sibling miners' in-cycle retries.

[ai-mountain]

@anderdc Could you please review this please? Thanks

[anderdc]

fix tests

[ai-mountain]

@anderdc I've fixed CI test failures.

[ai-mountain]

@anderdc Thanks.