Is your feature request related to a problem?

I want to write a DataTree as a netcdf file to a remote filesystem, either directly or via writing to in-memory bytes first. Right now this seems to be impossible.

• datatree.to_netcdf requires a filepath argument (no option for it to return bytes like dataset.to_netcdf())
• Passing a file-like object (e.g. io.BytesIO()) as the filepath results in ValueError: cannot save to a group with the scipy.io.netcdf backend, I believe it selects the scipy backend because it thinks (incorrectly) that's it's the only one capable of writing to a file-like object, but this backend then can't handle groups that are needed for DataTree
• datatree.to_netcdf(bytes_io, engine='h5netcdf') doesn't help (same error, specified engine is ignored by _dataset_to_netcdf and scipy selected instead)
• If I hack around the above (xr.backends.api.WRITEABLE_STORES['scipy'] = xr.backends.api.WRITEABLE_STORES['h5netcdf']), next obstacle is that H5NetCDFStore.open tries to read a magic number from the file-like object even when opening it in write mode: https://github.com/pydata/xarray/blob/main/xarray/backends/h5netcdf_.py#L166C9-L167C65
• After hacking around that (xarray.backends.h5netcdf_.read_magic_number_from_file = lambda *a,**kw: b"\211HDF\r\n\032\n"), I'm finally able to get datatree.to_netcdf(bytes_io) to work.
• Even then it still won't work with a file-like object that doesn't implement seek, but I believe this is a fundamental limitation of h5py.

Describe the solution you'd like

datatree.to_netcdf()   # Returns bytes like dataset.to_netcdf()

or

bytes_io = io.BytesIO()
datatree.to_netcdf(bytes_io)

or

with fsspec.open('remote-filesystem://path/to/file.nc', 'wb') as file_like_object:
  datatree.to_netcdf(file_like_object)

(although the latter may not be possible for file-like objects that don't implement seek.)

Describe alternatives you've considered

No response

Additional context

No response