Set Task Rl

The task's verifier in the workspace: a sealed RL task set of the workspace whose held-out test tasks decide the task's trained specialists and whose train tasks reinforcement fine-tuning trains them on after supervised fine-tuning; and the H100-hours each GRPO run of the task may take (none: no GRPO, reinforcement fine-tuning only). dataset_version_id null clears it. train_dataset_version_id: the RL runs' training set instead of the verifier's train tasks, a composite of RL task sets that keeps the verifier as a source marked solved at a stated solved_share (POST /datasets/{id}/versions), so RL on new tasks keeps the ones it solves.

The task's verifier in the workspace, its GRPO budget per run and its RL training set (null: the verifier's train tasks), and who set them, when (all null without one).