ENH: Replace pandas dependence/use with narwhals

[cluhmann]

Before

No response

After

No response

Context for the issue:

With the rise in popularity of packages such as polars, arrow, and others PyMC's dependence on pandas is looking less and less universally useful. Narwhals was developed for developers of python libraries that consumes dataframes who wish wishing to make their libraries completely dataframe-agnostic. So maybe we consider using that instead of pandas per se? Narwhals has no dependencies and has negligible overhead, so it seems relatively lightweight.

It will require some refactoring as it relies on (as subset of) the polars API.