set encoding parameters in addition to the original encoding

Is your feature request related to a problem?

When writing to disk with to_netcdf, the encoding argument causes existing encoding to be dropped. This is described in the docs.

What is a good approach to add encoding parameters in addition to the original encoding? e.g.

import rioxarray
import xarray as xr
import numpy as np

# make some random dummy netcdf file

data = np.random.rand(4, 4)
lat = np.linspace(10, 20, 4)
lon = np.linspace(10, 20, 4)

ds = xr.Dataset({"dummy": (["lat", "lon"], data)}, coords={"lat": lat, "lon": lon})

ds.rio.set_spatial_dims("lon", "lat", inplace=True)
ds.rio.write_crs("EPSG:4326", inplace=True)

# note the spatial_ref coordinate
print(ds.dummy)

<xarray.DataArray 'dummy' (lat: 4, lon: 4)> Size: 128B
...
Coordinates:
  * lat          (lat) float64 32B 10.0 13.33 16.67 20.0
  * lon          (lon) float64 32B 10.0 13.33 16.67 20.0
    spatial_ref  int64 8B 0

ds.to_netcdf("test.nc", mode="w")

# read it back in - ok

ds2 = xr.open_dataset("test.nc", decode_coords="all")

print(ds2.dummy)

<xarray.DataArray 'dummy' (lat: 4, lon: 4)> Size: 128B
...
Coordinates:
  * lat          (lat) float64 32B 10.0 13.33 16.67 20.0
  * lon          (lon) float64 32B 10.0 13.33 16.67 20.0
    spatial_ref  int64 8B ...

# now compress

ds2.to_netcdf("test_compressed.nc", mode="w", encoding={"dummy": {"compression": "zstd"}})

# read it back in - drops the spatial_ref

ds3 = xr.open_dataset("test_compressed.nc", decode_coords="all")

print(ds3.dummy)

<xarray.DataArray 'dummy' (lat: 4, lon: 4)> Size: 128B
...
Coordinates:
  * lat      (lat) float64 32B 10.0 13.33 16.67 20.0
  * lon      (lon) float64 32B 10.0 13.33 16.67 20.0

this is because rioxarray stores "grid_mapping" in the encoding.

so what is a nice generic way to specify encoding in addition to the original encoding?

encoding = ds2.dummy.encoding.copy()
encoding["compression"] = "zstd"
ds2.to_netcdf("test_compressed_2.nc", mode="w", encoding={"dummy": encoding})

ValueError: unexpected encoding parameters for 'netCDF4' backend: ['szip', 'zstd', 'bzip2', 'blosc']. Valid encodings are: ...

It seems not possible to pass the original encoding back in (even unmodified) due to additional checks

Describe the solution you'd like

in to_netcdf() be able to specify encoding in addition to the original encoding

Describe alternatives you've considered

No response

Additional context

No response

Start in xarray/backends/api.py at the additional encoding-parameter checks linked in the issue, then reproduce the rioxarray example with to_netcdf. The change is complete when callers can add compression settings without losing the original grid_mapping encoding or triggering invalid-backend-parameter errors.