fix: paginate PR file fetch for scoring accuracy

[taooutsider]

Summary

• paginate GET /repos/{owner}/{repo}/pulls/{number}/files using per_page=100 + page
• aggregate all changed files before scoring
• preserve existing retry/backoff behavior on transient failures

Why

GitHub returns only 30 files by default for PR file listings. For PRs with many changed files, scoring could miss file diffs and undercount contribution value.

Changes

• update get_pull_request_file_changes() to fetch all pages
• add tests for:
  • multi-page aggregation
  • retry when page 2 fails, restarting from page 1

Validation

• pytest -q tests/utils/test_github_api_tools.py -k FileChangesRetryLogic
• result: 10 passed

[taooutsider]

This fix prevents truncated PR file lists when changed files > 30 (GitHub default), which can under-score larger contributions. Happy to adjust if you prefer Link-header based pagination.