[Doc] Enhancement the document for dataset and s3

[mapleFU]

Describe the enhancement requested

Currently, we has the issue below:

1. [Python][Dataset][Parquet] Enable Pre-Buffering by default for Parquet s3 datasets #36765
2. [Python] Read table stuck and hangs forever #37139

When reading parquet from s3, it's important to prefetch some file or row-groups. However, larger prefetch depth might cause more memory usage. And what make it even worse is that, interfaces like to_table and other reading might has different behaviors for prefetch. So we'd better and more document for it.

Component(s)

Documentation, Parquet, Python