fix: coerce MirrorReviewSummary review counts to int

[e11734937-beep]

Summary

MirrorReviewSummary.from_dict is the only model parser that doesn't coerce its
numeric fields with int() — every other field does (additions, deletions,
commits_count, pr_number, …). maintainer_changes_requested_count feeds the
review-quality multiplier and is written to the analytics DB, so a string-typed
count from the mirror would silently break comparisons/arithmetic.

This wraps the four counts in int() to match the established convention.
Behaviour-preserving for well-formed payloads — a robustness/parity fix, not
a scoring-policy change (no weights/multipliers/thresholds touched).

Type of Change

• Bug fix

Testing

Added test_counts_coerced_to_int; uv run pytest tests/utils/test_mirror_models.py passes.

cc @anderdc @LandynDev

[anderdc]

These counts are serialized mirror-side via json_build_object, which emits JSON numbers — the validator always parses them as int, so a string-typed count can't reach here and the int() coercion is dead code. The sibling fields that coerce (additions/deletions/commits_count) are raw DB columns where the type can vary; these aren't. Closing.