docs(pr-test-sync): deploy check, affected-page runs, runtime data, parent-commit red proof, readonly merge order

[saqibmanan]

Updates .claude/skills/pr-test-sync/SKILL.md. The scheduled routine reads it from main.

Lessons that were in the local copy but never pushed (first commit): check collection under CI's real marker filter (regression is never selected), take the file number from the base branch, 429 rate-limit skips, and try changing -n before blaming a selector when a test fails only in CI.

New, from DataSpaceFrontend#476 / CivicDataSpace-test#122 (one commit each):

• Merged ≠ deployed. #476's deploy failed its gate and rolled back, so dev kept the old page. Check the deploy run on the merge SHA and look at the page in Playwright MCP before trusting a dev run.
• Affected pages. Resolve the real route param, find the existing tests through their Page Objects, run them, and check that they actually take the changed path. test_con_007 opens a use case with no dashboard.
• Runtime test data. Discover records through the API rather than hardcoding ids. Parametrize every case, including "nothing to show". Prefer records that force the transformation: a data edit mid-run made the Superset test pass without proving standalone=1 was added.
• Parent-commit red proof for frontend PRs. Serve the app locally against the dev API at the merge commit and at its parent. Includes the local gotchas: codegen, --webpack, no header nav with stub auth, Superset CSRF only on localhost.
• readonly merge order. The frontend and backend prod deploys both run readonly from CI. A test for a feature that is only on dev will roll back the next prod deploy, so check prod and say "merge after #N is on prod".