fix(validator): count only canonical valid solved issues

[Khaostica]

Summary

Move acc.valid_solved += 1 in _score_miner_issues past every credibility-only gate (same-account, one-issue-per-PR canonical, per-repo token threshold). Previously the increment fired on token-score alone, before the canonical check, so seven sibling issues closed by one qualifying PR could satisfy the per-repo MIN_VALID_SOLVED_ISSUES eligibility gate even though only one was scoreable.

Adds regression coverage for both the single-repo and per-repo (#1293) shapes, plus a credibility-only assertion on the same-account path, and updates two existing tests whose assertions/fixtures encoded the buggy "seven-counts-as-seven" behavior.

This is a rebased, refactor-aware re-do of #1273 (closed for unresolved conflicts after #1293 landed feat(validator): per-repository eligibility). The original patch was written against the pre-refactor global valid_solved_count / MIN_TOKEN_SCORE_FOR_BASE_SCORE; this one targets acc.valid_solved / cfg.min_token_score_for_valid_issue and also moves the increment after the token-threshold quality gate so below-threshold canonical issues no longer inflate the count.

Related Issues

Fixes #1269. Supersedes #1273.

Type of Change

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

Testing

• Tests added/updated
• Manually tested

Local verification:

• pytest tests/validator/issue_discovery/ — 56 / 56 pass
• ruff check gittensor/validator/issue_discovery/scan.py tests/validator/issue_discovery/test_scan.py — clean
• ruff format --check ... — clean
• Manually inspected the diff against the canonical-rule semantics in [Bug] Non-canonical one-issue-per-PR siblings still satisfy the valid-solved eligibility gate #1269 and the per-repo eligibility refactor in feat(validator): per-repository eligibility #1293.

Test changes:

• test_self_issue_counts_credibility_but_no_score — adds total_valid_solved_issues == 0 assertion (same-account is credibility-only, must not bump the gate).
• test_two_miners_shared_pr_only_earliest_scores — B now has total_valid_solved_issues == 6 (was 7, the bug). B stays eligible since MIN_VALID_SOLVED_ISSUES was dropped to 3 since [Bug] Non-canonical one-issue-per-PR siblings still satisfy the valid-solved eligibility gate #1269 was filed; the dedicated eligibility-gate regression lives in the new sibling test below.
• test_within_miner_one_issue_per_pr_still_holds — miner now has total_valid_solved_issues == 7 (was 8). Still eligible (7 unique canonical valid solved).
• test_all_mirror_miner_below_threshold_passes_spam — existing test that relied on the bug (eight issues sharing one PR all counted as valid_solved). Gave each issue its own solving PR and seeded all eight so the spam-mult assertion still has an eligible miner post-fix.
• New test_non_canonical_siblings_do_not_satisfy_valid_solved_gate — direct [Bug] Non-canonical one-issue-per-PR siblings still satisfy the valid-solved eligibility gate #1269 repro: seven sibling issues, one canonical PR → valid_solved == 1, ineligible (1 < 3), score == 0.
• New test_non_canonical_siblings_per_repo_eligibility — multi-repo variant: 7 unique canonical issues in repo A and 7 siblings on one shared PR in repo B → only repo A contributes to scored issues; total valid_solved == 8 (7 + 1 canonical sibling).

Checklist

• Code follows project style guidelines
• Self-review completed
• Changes are documented (if applicable)

[anderdc]

valid_solved is the can't-parse fallback for issues whose discovery was credited by a qualifying PR — it's intentionally broader than the score-row set, which is what the canonical one-issue-per-PR check governs. In practice a PR closes at most a handful of issues; when it closes more, the maintainer intake on the issues is the primary control, not the validator-side counter. Stripping siblings here drops eligibility credit on routine cross-miner collisions and batch-fix PRs (Fixes #X #Y #Z) where honest discoverers share a solving PR. Closing.