to_zarr removes global attributes in destination dataset

What happened?

Adding new variables to a zarr dataset with to_zarr() always removes the existing global attributes. New global attributes in the source dataset are not always added to the destination dataset depending on how to_zarr() is called.

What did you expect to happen?

I would expect that existing global attributes would always be preserved. If there are new global attributes I would expect them to be added to the existing global attributes instead of replacing all existing global attributes.

Minimal Complete Verifiable Example

import xarray as xr
from pyproj import CRS

local_zarr = 'sample.zarr'

ds_sample = xr.tutorial.load_dataset("air_temperature")

# Make a local copy
ds_sample.to_zarr(local_zarr, mode='w')
ds_sample = xr.open_dataset(local_zarr, engine='zarr', 
                            backend_kwargs={'consolidated':True}, chunks={}, 
                            decode_coords=True)

# Create CRS metadata
crs_meta = CRS.from_epsg(4326).to_cf()

ds_new = xr.Dataset(data_vars={"crs": ([], 1, crs_meta)})
ds_new.attrs['note'] = 'please add this'

# Add all variables from ds_new to the zarr
# NOTE: This adds the new global attribute but also removes
#       all existing global attributes
ds_new.to_zarr(local_zarr, mode='a')

# Add selected variable(s) to zarr dataset
# NOTE: This does not copy new global attributes 
#       and removes all existing global attributes
# ds_new['crs'].to_zarr(local_zarr, mode='a')

# Re-open local zarr store
ds_sample = xr.open_dataset(local_zarr, engine='zarr', 
                            backend_kwargs={'consolidated':True}, chunks={}, 
                            decode_coords=True)

ds_sample

MVCE confirmation

• Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
• Complete example — the example is self-contained, including all data and the text of any traceback.
• Verifiable example — the example copy & pastes into an IPython prompt or Binder notebook, returning the result.
• New issue — a search of GitHub Issues suggests this is not a duplicate.
• Recent environment — the issue occurs with the latest version of xarray and its dependencies.

Relevant log output

No response

Anything else we need to know?

No response

Environment

Start by running the minimal example in the issue and compare the destination attributes after both to_zarr() calls. Trace the relevant to_zarr behavior for dataset and data-array inputs, then verify that existing attributes remain and new attributes are merged in the resulting Zarr dataset.