interaction with `timedelta64`

Arrays with dtype timedelta64 are basically quantities with dimension [time], and they even have explicit units.

It would be really useful to be able to convert between those and pint quantities. The core logic should probably live in pint, but I think we should figure out a way to make the conversion easier:

arr = xr.Dataset(
    coords={
        "delta1": ("time", np.arange(100, dtype="timedelta64[s]")),
        "delta2": ("time", np.arange(100), {"units": "s"})
    }
).pint.quantify()
arr.pint.to({"delta1": "s", "delta2": "timedelta64[s]"})

(not sure if overloading .to would be worth it, but if not I'm sure we can come up with something better).

Similarly, I'd like to be able to infer the frequency of a datetime64 object and get the frequency (if any) as a quantity. It seems this works (found here):

from pandas.tseries.frequencies import to_offset

timedelta = to_offset(xr.infer_freq(time)).delta.to_numpy()

where the timedelta would then be converted to pint using the functionality described above.

Start by reviewing the accessor paths behind Dataset(...).pint.quantify() and .pint.to(), then inspect xarray.infer_freq and pandas.tseries.frequencies.to_offset. Done should cover the proposed timedelta64-to-quantity conversion and datetime64 frequency handling, but the issue leaves the API and implementation boundary open.