feat(gym): add actor checkpoint participant

[ananthsub]

Summary

• discover checkpoint-capable Gym model, agent, and resources participants from live capabilities
• coordinate deadline-bounded prepare, commit, restore, resume, abort, and status operations inside the NemoGym Ray actor
• freeze actor-local rollout membership and acknowledge exact server-issued completion receipts before releasing delivered results
• validate aggregate manifest identity and integrity while keeping checkpoint payloads opaque to NeMo RL
• pin NeMo Gym to generation-safe stack head 91ea76cec from Gym#2946

This is the second PR in the stack and depends on RL#4085. It intentionally excludes SingleController orchestration.

Validation

• focused actor and existing NemoGym tests against the final Gym pin: 65 passed
• actor-focused suite before the final pin: 73 passed
• integrated Gym contract tests: 3 passed
• Gym generation-safe stack: 624 passed
• Ruff and formatting: passed
• focused Pyrefly: 0 errors

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[github-actions]

❌ Submodule Fast-Forward Check Failed

Check based on commit: e7e58f5 (PR #4088 from ananthsub/gym-checkpoint-actor)

❌ Submodules that need attention:

Gym: ❌ Commits have DIVERGED from a common ancestor
TARGET (ananthsub/gym-rollout-checkpoint-identity branch): https://github.com/NVIDIA-NeMo/Gym/commits/fd5e84d6b1c485c80e7ae61553bbd485611c03b4/
CURRENT (PR #4088 from ananthsub/gym-checkpoint-actor): https://github.com/NVIDIA-NeMo/Gym/commits/91ea76cec2eb150d56548d46ffbdfbd707805972/

Please ensure all submodule commits are fast-forwards of the ananthsub/gym-rollout-checkpoint-identity branch before merging.

[ananthsub]

Superseded by a GitHub Stack whose branches are owned by NVIDIA-NeMo/RL rather than a fork.

[ananthsub]

Replacement upstream-owned GitHub Stack: #4092 → #4093.