Support serializing to json in a format similar to pandas "split" orientation

[CaselIT]

Description

Pandas supports a json serialization that's quite compact compared to the classic list of dict one.
This may be useful when sending dataframe via api calls.

Example:

df = (
    pl.DataFrame({k: [i for i in range(4)] for k in "abc"})
    .select(pl.col.a / 3, pl.col.c.cast(pl.String))
    .to_pandas()
)
print(df.to_json(orient="split"))

The above prints

{"columns":["a","c"],"index":[0,1,2,3],"data":[[0.0,"0"],[0.3333333333,"1"],[0.6666666667,"2"],[1.0,"3"]]}

Since polars has no index, it could output just columns and data.
Note that data, a bit unfortunately, is row-major, not column-major.

While using to_pandas() to get a pandas dataframe is an option, it's usually quite slow, and seems a waste to do that just to then serialize it again.