fix(scoring): skip base_ref gate when mirror omits base_ref

[Mahdi3Bani]

Summary

check_merged_branch_eligibility gated the base-ref check on base_ref or ''. When the mirror omits base_ref, that becomes '', which never matches the acceptable branch set — so the PR is rejected at load time by _should_skip_merged_mirror_pr, never enters merged_prs, and earns no OSS contribution score.

Pre-backfill mirror rows legitimately carry no branch metadata. MirrorPullRequest declares base_ref: Optional[str] and populates it with data.get('base_ref'), so None is an expected runtime value, not a corrupt row.

Two places in the codebase already treat missing branch metadata as fall-through rather than grounds for rejection:

• Issue discovery — issue_discovery/scan.py gates on sp.base_ref is not None, with a comment stating that a missing base_ref means pre-backfill data and should fall through. Covered by test_missing_base_ref_falls_through_to_solved.
• This same function — the head_ref check falls through on missing head_ref / head_repo_full_name, and the docstring documents that as intended behavior.

The base_ref gate was the odd one out. This changes it to gate on presence, restoring parity across the two paths. An explicitly malformed (empty-string) base_ref is still rejected — only genuinely absent metadata falls through.

Also widens base_ref on the _pr test helper from str to str | None. It had drifted from the model, which is why the null case had no coverage — the helper couldn't express it. head_ref was already str | None there.

Related Issues

Fixes #1636

Type of Change

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

Testing

• Tests added/updated
• Manually tested

Added test_null_base_ref_skips_check, mirroring the existing test_null_head_ref_skips_check. Confirmed it fails on test without the source change and passes with it.

Full suite: 893 passed. pre-commit run --all-files and --hook-stage pre-push (ruff, pyright, vulture, pytest) all pass.

No CLI output changes.

Checklist

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

[Mahdi3Bani]

#1641 predates this and makes the same change to the base_ref gate — I missed it before opening this one, apologies for the duplicate.

The difference is test coverage: this adds test_null_base_ref_skips_check, and widens base_ref on the _pr test helper from str to str | None. The helper had drifted from MirrorPullRequest.base_ref: Optional[str], which is why the null case had no coverage — pyright rejects the test without that change.

Happy to close this in favour of #1641 if you'd rather take that one. Whichever lands, I'd suggest the regression test comes with it.