Add a "why nested-pandas?" section to the docs

[dougbrn]

Nested-Pandas is an enhancement to pandas that has some degree of overlap with other APIs (pandas multi-index, ragged arrays). It would be nice to have a section of the docs that explains the motivation behind nested-pandas and the strengths of it compared to the existing offerings for pandas users. Multi-indices is the main one (with our obvious strengths being that we can store an arbitrary number of columns at a given level with no data duplication, whereas multi-indices are limited to just using the index).

[hombit]

The list of alternatives I can think about, including what @dougbrn mentioned in the issue:

1. Two table approach with join and group-by, it is already covered in the performance notebook.
2. Single flat table being group-by. It is huge!
   a. Multi-index, but it doesn't support multiple "nested" columns.
   b. Group-by some column, but it is slow.
3. Array-columns (same as list columns, "ragged" arrays), e.g. a "vector or vector references". No user-interface benefits (which columns are logically connected?) and performance is bad. We can also analyze (de)serialization performance with formats like HDF5 and FITS

I think we can add a documentation page/section where we give a short summary of all these alternatives with some ""anecdotal" benchmark numbers, while providing the benchmark code separately.

We also should tell that nested-pandas doesn't invent a new format or data type for arrow or parquet, and that our pandas dtype is compatible with what pandas already has. However, it gives a user interfaces missing from pandas and other data-analysis libraries like polars or DuckDB