fix(issue-discovery): honor per-repo min_token_score in solving-PR cache

[galuis116]

Summary

PR #1330 plumbed min_token_score_for_base_score through the OSS scoring path (mirror/scoring.py) and the issue-discovery solving-PR cache-MISS path (scan.py:676-680), but the solving-PR cache-population path at scan.py:409 kept reading the global MIN_TOKEN_SCORE_FOR_BASE_SCORE constant — silent state asymmetry between the cache and the path that fills its misses.

With a repo-level override set, OSS gives a merged PR with token_score = 3 a non-zero base_score (post-#1330), but _build_solving_pr_cache drops the entry because 3 < 5. Every sibling discoverer's lookup for that PR then cache-MISSes and refetches via MirrorClient.get_pr_files; on a MirrorRequestError or scoring_data_stored = False the per-repo accumulator sets fetch_failed = True, and the discoverer loses both valid_solved credit and discovery_earned_score for that issue — even though the data was already in memory on this validator round. No repo in master_repositories.json currently sets the override, so the bug is latent, but the moment an admin uses the field the cache and the cache-MISS path disagree about which PRs are scorable. Same forward-looking-consistency posture as #1330.

The docs are explicit on the contract: every eligibility constant "is a global default that any repository can override in its config entry" (docs.gittensor.io/oss-contributions.html §"Per-repo configs vary"), and the cross-miner cache is specified as a pure pass-through — "Miners' own PRs are pre-scored during OSS evaluation and reused for free" (docs.gittensor.io/issue-discovery.html §"Base Score Calculation"). The current cache filter breaks the "reused for free" promise whenever the override is in effect.

Closes #1355.

Fix

• _build_solving_pr_cache now takes mirror_repos: Dict[str, RepositoryConfig] and resolves the per-PR threshold via resolve_eligibility(repo_config.eligibility).min_token_score_for_base_score, falling back to the global MIN_TOKEN_SCORE_FOR_BASE_SCORE when the repo isn't in mirror_repos (defensive — in practice every merged_prs entry is from a tracked repo).
• The single call site in run_issue_discovery (scan.py:165) now passes mirror_repos.
• Docstring updated to spell out the parity with the cache-MISS path and the OSS scoring path.

Test plan

• New regression tests in tests/validator/issue_discovery/test_scan.py::TestSolvingPrCache:
  • test_per_repo_override_keeps_pr_above_repo_threshold — a repo with min_token_score_for_base_score = 2 keeps a PR whose token_score = 3 (clears override, fails global default).
  • test_untracked_repo_falls_back_to_global_threshold — defensive: a repo missing from mirror_repos still gates on the global default, doesn't crash.
• Existing TestSolvingPrCache tests (test_build_cache_from_multiple_miners, test_cache_first_occurrence_wins_on_duplicate, test_below_threshold_prs_excluded_from_cache) updated to pass mirror_repos; behavior preserved when no override is configured.
• uv run pytest tests/validator/issue_discovery/test_scan.py -q — 66 passed.
• uv run pytest tests/validator/ -q — 519 passed.
• uv run ruff check gittensor/validator/issue_discovery/scan.py tests/validator/issue_discovery/test_scan.py — clean.
• uv run ruff format --check gittensor/validator/issue_discovery/scan.py tests/validator/issue_discovery/test_scan.py — clean.
• uv run pyright gittensor/validator/issue_discovery/scan.py tests/validator/issue_discovery/test_scan.py — 0 errors, 0 warnings.

Files changed

• gittensor/validator/issue_discovery/scan.py — _build_solving_pr_cache accepts and uses the resolved per-repo threshold; call site in run_issue_discovery updated to pass mirror_repos.
• tests/validator/issue_discovery/test_scan.py — 2 new regression tests; 3 existing TestSolvingPrCache tests updated to pass mirror_repos; RepoEligibilityConfig added to the existing load_weights import block.