fix: add pagination, error handling, and modernize runtime

[peterkrauz]

Summary

• Paginate GitHub API responses — follows Link: rel="next" headers, fetching up to 1,000 comments (10 pages × 100). Previously only the first 30 were fetched, causing false negatives on active PRs.
• Handle HTTP errors — raises with status code and response body excerpt instead of silently parsing error JSON as comment data.
• Modernize runtime — Python 3.7 (EOL since June 2023) → 3.13, requests 2.27.1 → 2.32.3, removed unused curl/jq from the Docker image.

Other improvements

• Exit code protocol replaces fragile "0" stdout sentinel
• Failure output goes to stderr as human-readable text instead of a Python list repr
• set -euo pipefail in entrypoint
• Full type annotations on all functions
• Removed unused imports (logging, dumps)
• README updated: version ref @v0, removed unnecessary container: line, mentions pagination

Test plan

• Create a release on a fork and verify the publish workflow succeeds
• Trigger the action on a PR with <30 comments — should behave as before
• Trigger the action on a PR with >30 comments — should now find users beyond page 1
• Verify failure output is readable in the Actions log
• Verify a revoked/invalid token produces a clear RuntimeError instead of a crash