Handling of cftime.datetime objects in HydroMT

[hboisgon]

Working with DatetimeNoLeap calendars, I found this is not so easily handled with HydroMT, especially for xarray objects.

For example, in xarray.open_dataset there is a use_cftime flag which is None by default. This means that times will be decoded to np.datetime64[ns] where possible and else to cftime.datetime:
https://docs.xarray.dev/en/stable/generated/xarray.open_dataset.html

And as these two objects are quite different (ie not comparable), this leads to quite some exceptions.
For example in HydroMT, we have a preprocess functions to convert cftime to datetimeindex but by doing this, we loose the calendar information (for example from 'noleap' to 'proplectic_gregorian').

hydromt/hydromt/data_adapter/data_adapter.py

Lines 25 to 28 in 4784272

	def to_datetimeindex(ds):
	if ds.indexes["time"].dtype == "O":
	ds["time"] = ds.indexes["time"].to_datetimeindex()
	return ds

Also, in get_rasterdataset, we put an exception to only select the correct times based on time_tuple only if time is of type np.datetime64, which means that data with cftime.datetime type are not cut in time:

hydromt/hydromt/data_adapter/rasterdataset.py

Lines 246 to 251 in 4784272

	# clip tslice
	if (
	"time" in ds_out.dims
	and ds_out["time"].size > 1
	and np.issubdtype(ds_out["time"].dtype, np.datetime64)
	):

I think there is more issues downstream with the type cftime.datetime which is why we put the preprocessor in but would need to investigate.

So I wonder if like xarray we keep the two types possible and if so we should better test and check that every functionality still works for both types of time index.