[cfg, megatron, doc] fix: honor actor.router_replay or fail if it disagrees with engine

[YeonwooSung]

What does this PR do?

Fixes #7463.

actor.yaml and the Ascend NPU guide document actor_rollout_ref.actor.router_replay.mode. The Megatron/VeOmni worker only reads actor.{megatron,veomni}.router_replay.mode. Setting the documented key left routing replay disabled.

This PR:

• Copies a non-disabled actor-level router_replay onto the engine config when the engine is still disabled.
• Raises if both copies are set to different non-disabled modes.
• Updates the NPU guide / parameter table to name the engine key as canonical, with the top-level key as a compatibility alias.
• Makes experimental/separation R2 vs R3 handling read the engine copy first.

Closed #3762 asked whether routing replay would be supported (it is). This is the split-schema no-op, not a request to add the feature.

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

Local machine did not have the full verl test env. Ruff passed on the touched files. Please run the CPU test in CI.

API and Usage Example

# documented key now enables engine-side replay when the engine key is still disabled
actor_rollout_ref.actor.router_replay.mode=R3
actor_rollout_ref.rollout.enable_rollout_routing_replay=True

# still valid, and preferred
actor_rollout_ref.actor.megatron.router_replay.mode=R3

Design & Code Changes

McoreActorConfig / VeOmniActorConfig.__post_init__ call ActorConfig.sync_router_replay_to_engine(). Engine router_replay fields are frozen, so the copy uses object.__setattr__ (same pattern as FSDPActorConfig syncing strategy).

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 added under tests/workers/config/.
• 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 draft the patch. A human author reviewed every changed line.