Improve support for non-numpy dtypes

[hoxbro]

We should improve the support for different dtypes; currently, we really only support numpy.

dtypes which make sense to support:

• Pandas dtype, like pd.Int64dtype
• Pyarrow, like date32[arrow] (This can give problems see Exceptions raised when trying to create stacked bar chart with continous x-axis. #6302)
• Polars dtype, this will likely be done in Add first-class support for polars #5939.

[hoxbro]

Another examples is inspect_polygons does not seem to like string[pyarrow] (reduced example of NYC buildings)

import dask
import holoviews as hv
import spatialpandas.io as sio
from holoviews.operation.datashader import datashade, inspect_polygons

hv.extension("bokeh")
dask.config.set({"dataframe.convert-string": False})  # False works

# http://s3.amazonaws.com/datashader-data/nyc_buildings.parq.zip
ddf = sio.read_parquet_dask("./data/nyc_buildings.parq").persist()

ddf.dtypes  #  type and name is `pyarrow[string]` or object.

polys = hv.Polygons(ddf)
shaded = datashade(polys)
hover = inspect_polygons(shaded).opts(fill_color="red", tools=["hover"])
(shaded * hover).opts(width=800, height=800)

Edit: It does not seem to be a problem with non-numpy dtype directly.