Cast kernel between int32, date and string for partition columns

[alippai]

Describe the enhancement requested

Dataset filters on partition values can be hard to define using pyarrow.

Eg. type=0/date=2023-01-01/0.parquet needs filters ("type", "=", 0) and ("date", "=", "2023-01-01").
Can Arrow support ("type", "=", "0") and ("date", "=", datetime.date.today()) as well? My understanding is that these need only missing equal (lt, gt) kernel implementations.

I had the errors using pyarrow 11.0.0

Component(s)

C++, Python