via lambda doesn't seem to make much of a difference. In both cases,
when doing just a selection, it'll be kept lazy, but say rescaling by
multiplying by a scalar leads to the data being loaded.

Probably something where dask comes in.

Since the backend is registered by means of pyproject.toml now

No need to have a psc.RunInfo until later

This is all about gettings things more in line with other xarray backends

In particular this requires the subtlety of actually returning
xr.Variable from open_store_variable(), not xr.DataArray

keep compatibility with files missing this information, though, even
though there's not much we can do, other than special casing psc so
that at least doesn't break

however, keep current behavior, ie always reverse

but also keep reversing for pfd* tfd* to maintain compatibility

This isn't consistent with xarray typing, and easy enough to avoid
by sticking with open_dataset(Adios2Store.open(group))

Looks like just something left over from the template

We might not want to do this in the long run, since this leads to a
transitive dependency on adios2, which is binary and potentially iffy to
install, so if one were to want to use pscpy without the adios2 reader,
that could cause some hassles.