Add support for modin to improve apply operation performance

[Beforerr]

I think for most part we can just replace

import pandas as pd

with

# Conditional import
try:
    import modin.pandas as pd
    from modin.pandas import DataFrame, Series
except ImportError:
    import pandas as pd
    from pandas import DataFrame, Series

[shaypal5]

Ummm. I don't think that's something that should be done by default, with no way to control it with configuration.

Are we to assume anyone that installed modin prefers to have it replace pandas anytime, anywhere? I don't know. Maybe.

Do you have any reference to other similar libraries built on top of pandas and making the same change?

[Beforerr]

One library I know is swifter (See jmcarpenter2/swifter/issues/93). But you are definitely right, we should have a way to control it with configuration.

[Beforerr]

Or we may do in another way, by not raising TypeError in _transform, and everything hopefully should work smoothly.

try:
    import modin.pandas as mpd
except ImportError:
  pass 

class ApplyToRows(pdp.ApplyToRows):
    ...
    def _transform(self, X, verbose):
        ...
        if  isinstance(new_cols, pd.Series) or isinstance(new_cols, mpd.Series):
            ...
        if isinstance(new_cols, pd.DataFrame) or isinstance(new_cols, mpd.DataFrame):
            ...

[shaypal5]

I still think there should be a way to opt-in, not opt-out, into this change.

And then maybe it could be applied everywhere.

Look at the way libraries do it with global flags which are read once and can be set on import, or by configuration files an environment variables.