expose CFMaskCoder

Is your feature request related to a problem?

We have CFMaskCoder that lazily masks out the values of the _FillValue and missing_value attributes (if present).

For the Zarr backend, we have use_zarr_fill_value_as_mask which controls whether the fill_value attribute is masked: https://docs.xarray.dev/en/stable/generated/xarray.open_zarr.html

If True, use the zarr Array fill_value to mask the data, the same as done for NetCDF data with _FillValue or missing_value attributes. If False, the fill_value is ignored and the data are not masked. If None, this defaults to True for zarr_version=2 and False for zarr_version=3.

We have requests for masking data outside the [valid_min,valid_max] or the valid_range range: https://github.com/pydata/xarray/issues/8359.

Describe the solution you'd like

I propose we expose

coders.MaskCoder(attributes:Iterable[str], valid_range:bool)

(no CF) to allow configurable masking.

It may take some work to figure out how to set a per-backend encode/decode pipeline.

Start by reading the existing CFMaskCoder behavior and the open_zarr masking option described in the issue. Trace how backend encode/decode pipelines are configured, then determine how configurable masking and valid-range handling would fit each backend. Done means the proposed MaskCoder is exposed and the relevant backend pipelines support the requested masking choices.