Improve `Create a dataset` tutorial

Our [tutorial on how to create a dataset](https://huggingface.co/docs/datasets/create_dataset) is a bit misleading.
1. In **Folder-based builders** section it says that we have two folder-based builders as standard builders, but we also have similar builders (that can be created from directory with data of required format) for `csv`, `json/jsonl`, `parquet` and `txt` files. We have info about these loaders in separate [guide for loading](https://huggingface.co/docs/datasets/loading#local-and-remote-files) but it's worth briefly mentioning them in the beginning tutorial because they are more common and for consistency. Would be helpful to add the link to the full guide.
2. **From local files** section lists methods for creating a dataset from in-memory data which are also described in [loading guide](https://huggingface.co/docs/datasets/loading#inmemory-data).

Maybe we should actually rethink and restructure this tutorial somehow.