Megatron Bridge TrainingConfig support

**Is your feature request related to a problem? Please describe.**

[Here](https://github.com/NVIDIA-NeMo/RL/blob/main/nemo_rl/models/policy/megatron_policy_worker.py#L644) only `micro_batch_size`, `global_batch_size`, and `train_iters` are supported but other params may be needed too.

i.e. in async GRPO the `decrease_batch_size_if_needed` may be useful since some of GPUs don't participate in train loop

**Describe the solution you'd like**

That's a design decision with many options:

- leave it as is - I assume current logic is there to support train params for both FSDP and Megatron and it may not be a big deal to support all megatron bridge params
- add every TrainingConfig param as `train_`
- make `TrainingConfig` a part of `policy` or `policy.megatron_cfg` - I think it's the cleanest solution but deprecates previous configs