feat(gym): persist stable rollout attempt identity

[ananthsub]

Summary

• assign one stable logical _ng_rollout_id to each prompt sibling
• allocate numeric _ng_attempt_index values only when a sibling is dispatched
• preserve attempt allocation across partial row redispatch, outer retries, and rollout recovery
• derive Gym capture keys as <logical> for attempt 0 and <logical>-a<N> for later attempts
• bump rollout recovery state to schema v3 and reject unmigratable UUID-based v2 state with guidance

This is the first PR in the NeMo Gym checkpoint integration stack. It intentionally excludes the NemoGym checkpoint participant and SingleController orchestration. The dependent actor PR will pin the matching Gym stack and add cross-repository checkpoint lifecycle tests.

Related Gym tracking issue: NVIDIA-NeMo/Gym#3024

Validation

• focused rollout identity, manager, recovery, failure, redispatch, and NemoGym actor tests: 202 passed
• native redispatch suite: 25 passed
• capture-key compatibility and receipt lookup tests: passed
• Ruff and formatting: passed
• Pyrefly: 0 errors (run with the shared complete checkout because isolated worktree dependency submodules are absent)

Stack

Dependent actor participant: RL#4088.

[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.

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Superseded by a GitHub Stack whose branches are owned by NVIDIA-NeMo/RL rather than a fork.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