Enhance terminology in GRPO code

[nouhadziri]

The grpo_vllm_thread_ray_gtrl.py uses --actor_num_gpus_per_node to refer to the training processes which is a bit confusing and instead we can use --num_learners_per_node.

[vwxyzjn]

Also I think number_samples_per_prompt should be renamed to num_samples_per_prompt for consistency