libviewer for all

[lhoestq]

enable libviewer (page pruning features based on page index) for all datasets:

• export metadata using libviewer, which enables the use of the page index (pyarrow misplaces it somehow)
• use page pruning in /rows when a page index is present

this will make /rows faster and more efficient, and will enable cool use cases later like stopping re-converting datasets that are already in Parquet no matter their row group sizes