Docs: use relative path for preprocess_data_for_megatron.py

[Saibabu7770]

This updates GPT training docs to use a relative path:
python scripts/nlp_language_modeling/preprocess_data_for_megatron.py

Fixes #15130

[pzelasko]

Closing this PR as "won't merge".
The following collections have been moved to separate repos in https://github.com/NVIDIA-NeMo organization: avlm, llm, multimodal, multimodal-autoregressive, vlm, speechlm, diffusion.
If you still wish to proceed with this contribution, please re-open it in the relevant repo.