Add simple polars support

[hoxbro]

resolves #1023

This adds a simple Polars support to hvplot by converting it to a Pandas DataFrame before putting it into the machinery of hvplot. I'm trying to convert only the necessary columns for the DataFrame and LazyDataFrame.

Using the API as described here.

import hvplot.polars  # noqa
import polars as pl
from bokeh.sampledata.autompg import autompg

df = pl.DataFrame(autompg)
s = df.select("mpg").to_series()
ldf = df.lazy()

type(df), type(s), type(ldf)

df.hvplot(y="cyl")
s.hvplot()
ldf.hvplot.scatter()

[itsgifnotjiff]

Thank you so much Mr. @hoxbro . What do you mean by simple? Is there something that a Pandas DataFrame backed HvPlot/Holoviews object can fundementally do that a Polars one can not?

[philippjfr]

What do you mean by simple? Is there something that a Pandas DataFrame backed HvPlot/Holoviews object can fundementally do that a Polars one can not?

This PR simply converts the data to a pandas DataFrame before passing it to the existing codepaths. The longer term and more efficient approach will be to implement a data backend for HoloViews that natively understands and works with polars dataframes, avoiding any copies that occur in the polars -> pandas conversion and potentially executing specific computations, e.g. histograms, aggregations etc., directly with polars.

[itsgifnotjiff]

Thank you Mr. @philippjfr . I have a Panel app that uses Polars to load 12 GB SQLite and cast it to a Pandas DataFrame and I always wondered how it would perform using Polars alone.

[maximlt]

Except a few minor changes this looks good to me. This is going to make polars users very happy! I wish we had a place where to expose that more in the docs, the front-page isn't the best place.

Note to self, the release notes should mention that this is basic support and the data is casted to Pandas objects. By the way @hoxbro Series don't need to be casted?