verl E2E Testing

Validate a Megatron-Bridge change through verl's Megatron backend. This catches integration issues that Bridge-only tests miss: provider/configuration finalization, HF import through Bridge, PEFT wrapping, DDP wrapping, optimizer setup, rollout/ref wiring, and checkpoint ownership by an external RL loop.

Use this as an external compatibility smoke test after the relevant Bridge unit and functional tests are green.

This is not a replacement for Bridge correctness tests. The default verl run proves that the changed Bridge path can survive an external RL training loop; numerical parity, import/export fidelity, and feature-specific behavior still come from focused Bridge tests.

Scope

Think in coverage levels. Start with Level 0 and add only the levels justified by the change.