ResponseDataset Config Missing Required Fields in SFT Documentation

**Describe the bug**

To use local dataset to run SFT on a model, I followed instruction on[ this page](https://docs.nvidia.com/nemo/rl/latest/guides/sft.html#supervised-fine-tuning-in-nemo-rl) and decided to use [ResponseDataset class ](https://docs.nvidia.com/nemo/rl/latest/guides/sft.html#datasets). I followed the template and replaced my data config to
```
data:
dataset_name: ResponseDataset
train_data_path: # e.g., /path/to/local/dataset.jsonl or hf_org/hf_dataset_name (HuggingFace)
val_data_path:
input_key: , default is "input"
output_key: , default is "output"
train_split: , default is None # used for HuggingFace datasets
val_split: , default is None # used for HuggingFace datasets
```

however, when running the script I kept getting errors because the highlighted fields are required.

**Steps/Code to reproduce bug**

set data config in examples/configs/sft_openmathinstruct2.yaml to:
```
dataset_name: ResponseDataset
train_data_path: "/home/ameliay/RL/nemo_rl/data/datasets/sft_fraud_datasets/nss_to_rl_dataset_20251022_060530_train.jsonl" # e.g., /path/to/local/dataset.jsonl or hf_org/hf_dataset_name (HuggingFace)
val_data_path: "/home/ameliay/RL/nemo_rl/data/datasets/sft_fraud_datasets/nss_to_rl_dataset_20251022_060530_val.jsonl"
input_key: "problem"
output_key: "expected_answer"
```

**Expected behavior**

The expected fields should be specified in the instruction here:

**Additional context**

Add any other context about the problem here.