API: Add Calendar Day Frequency ("CD")

This is a proposal to replace #20633.

offsets.Day ('D') is documented to represent calendar day; however, Day arithmetic and usage of the 'D' offset alias (e.g. df.resample('D'), timedelta_range(..., freq='D'), etc.) currently respects absolute time (i.e. Day acts like Timedelta(days=1)). This usage is ingrained pretty deeply in a lot of methods and operations and is fairly difficult to walk back to respect the notion of calendar day as stated in the docs.

Instead, I propose to keep Day as is (a timedelta-like frequency like Hour ('H')) and add a new frequency CalendarDay, 'CD'. It would act very similarly to DateOffset(day=1) but act like a frequency. Thoughts?

[chris-b1]

Only partially related, but I’ve also thought it would be nice to have frequencies for other DateOffset cases, e.g. a calendar year like the below.

In [7]: pd.date_range('2014-01-15', periods=10, freq=pd.DateOffset(years=1))
Out[7]: 
DatetimeIndex(['2014-01-15', '2015-01-15', '2016-01-15', '2017-01-15',
               '2018-01-15', '2019-01-15', '2020-01-15', '2021-01-15',
               '2022-01-15', '2023-01-15'],
              dtype='datetime64[ns]', freq='<DateOffset: kwds={'years': 1}>')

So, maybe a whole set of 'CD', 'CM', 'CA'?

[mroeschke]

Closing in favor of further discussion in #22864 (especially about deprecation procedure)