Added crucial comment in fine-tune-embedding-model-for-rag.ipynb

[ChristianBernhard]

• added a valuable comment, as one codesnippet was really confusing, as we used the test and train-splits for the document pool but we have to treat both as train splits, when loading from JSON, because there wouldn't be a predefined split called test. Summary: The code is correct but it looks like a typo when having two variables: test_dataset and train_dataset but both use the split train