ci: keep nightly suites running after a test fails

[Rockdu]

What

• Nightly (schedule) runs pass --continue-on-error to every run_suite.py suite, so a failing test no longer stops the rest of its suite.
• New workflow_dispatch input continue_on_error (default false) lets a manual run do the same.

Why

Each suite stops at its first failing test. Since the 2026-09-27 nightly, test_krea2_nft_ocr_2xGPU has failed in stage-c-3-gpu-h200, so test_sd3_ocr_grpo_2xGPU, which runs after it, has not run on any nightly since. A nightly exists to report every drift at once; fail-fast hides everything behind the first failure.

The job still fails: run_unittest_files marks the suite failed on any failing test, and --continue-on-error only skips the break.

Files

• .github/workflows/pr-test.yml: the new dispatch input, the flag on all six suite commands, and the header note on nightly behaviour.

Checklist

• pre-commit run --all-files passes — not run
• Added/updated tests for new behaviour — no tests: workflow-only change; the workflow YAML parses and all six suite jobs carry the flag
• pytest -x is green — not run, no Python changed
• If launch flags changed, python3 train.py --help still parses — n/a
• If a public flag was added, it appears in the CLI reference docs — n/a, no CLI flag added
• If an example was added, it has a real walkthrough — n/a