FR: dtype for Decimal data

[tswast]

Currently, if one wants to store decimal.Decimal data, the pandas object dtype must be used.

Alternatively, one could use one of the pyarrow decimal types with fletcher https://github.com/xhochy/fletcher

I'm opening this issue to investigate how a (db)decimal dtype(s) might work and what the benefits would be beyond the obvious performance improvement potential.

[tswast]

There's already a DecimalDtype defined in tests.

https://github.com/pandas-dev/pandas/blob/ae5b2b7bab84a9c276ca14d0a46060735a0bd601/pandas/tests/extension/decimal/array.py#L38

Added 4 years ago in pandas-dev/pandas#19520

I imagine we'd want something backed by pyarrow these days, as there isn't really a natural numpy type, especially if we wanted to support NULL values.

[tswast]

Possibly redundant with pandas ArrowDtype + arrow decimal128 / decimal256, as these can unambiguously describe BigQuery's NUMERIC and BIGNUMERIC types, respectively.