[MLPerf 6.1][GRPO] Update Qwen3.5 validation and timeout rules

[jepio]

Summary

• Update the qualified Qwen3.5 validation set from 256 tasks to 251 tasks and the pass@4 rollout count from 1024 to 1004.
• Fix the training and validation agent timeout at 1,800 seconds.
• Fix the training and validation test and command timeouts at 60 seconds.

Rationale

Dataset validation removed five invalid validation tasks. The updated NeMo-RL reference also finalizes the runtime limits that affect rollout and reward behavior.

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[ShriyaRishab]

Approving based on task force email threads.