[cfg, megatron, doc] fix: drop unused actor.router_replay in favor of engine config - #7466

What does this PR do?

Fixes #7463.

actor.yaml exposed actor_rollout_ref.actor.router_replay, but Megatron/VeOmni workers only read actor.{megatron,veomni}.router_replay. The documented top-level key was a silent no-op.

Per review, this PR drops the unused top-level key instead of aliasing it:

• Remove router_replay from actor.yaml and from ActorConfig.
• Keep the live engine field on actor.megatron.router_replay / actor.veomni.router_replay.
• Point the Ascend NPU guide and parameter table at those engine keys only.
• Read the engine copy in experimental/separation when deciding R2 vs R3.

actor_rollout_ref.actor.router_replay.mode=... now fails in Hydra struct mode. The supported override is actor_rollout_ref.actor.megatron.router_replay.mode=R3 (or veomni).

Closed #3762 asked whether routing replay would be supported (it is). This issue is the leftover dead key.

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: https://github.com/verl-project/verl/pulls?q=is%3Apr+router_replay+actor.router_replay
• Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)

Test

pytest tests/workers/config/test_actor_router_replay_sync_on_cpu.py -q

Ruff passed on the touched Python files. Please run the CPU test in CI.

API and Usage Example

# Megatron
actor_rollout_ref.actor.megatron.router_replay.mode=R3
actor_rollout_ref.rollout.enable_rollout_routing_replay=True

# VeOmni
actor_rollout_ref.actor.veomni.router_replay.mode=R3

Design & Code Changes

The worker already selected self.config.actor.megatron.router_replay / self.config.actor.veomni.router_replay. Removing the unused actor-level schema avoids a second source of truth.

ref.router_replay in ref.yaml is unchanged; this review only asked to drop the actor-level key.

Checklist Before Submitting

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: CPU unit tests assert ActorConfig no longer has a top-level router_replay field and that the engine field still works.
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace.
• If your PR is related to the recipe submodule, please also update the reference to the submodule commit via git submodule update --remote or cd recipe && git pull origin main.

AI assistance was used to locate the bug and apply the review. A human author should review every changed line.