Add `xarray.open_dataset` backend

https://docs.xarray.dev/en/stable/internals/how-to-add-new-backend.html describes how to add a new backend to xarray's open_dataset. Adding a backend will make it easier for users to consume kerchunk references.

Here's a prototype:

from xarray.backends import BackendEntrypoint


class KerchunkBackendEntrypoint(BackendEntrypoint):
    def open_dataset(
        self,
        filename_or_obj,
        *,
        drop_variables=None,
        **kwargs,
    ):
        return my_open_dataset(filename_or_obj, drop_variables=drop_variables, **kwargs)

    open_dataset_parameters = ["filename_or_obj", "drop_variables"]

    def guess_can_open(self, filename_or_obj):
        try:
            _, ext = os.path.splitext(filename_or_obj)
        except TypeError:
            return False
        return ext in {".json", ".parquet"}

    description = "Use Kerchunk files in Xarray"
    url = "https://fsspec.github.io/kerchunk/"

def my_open_dataset(filename_or_obj, drop_variables, **kwargs):
    m = fsspec.filesystem("reference", fo=filename_or_obj, **kwargs).get_mapper()
    return xr.open_dataset(m, engine="zarr", consolidated=False)

This would be usable like

ds = xr.open_dataset(refs, engine="kerchunk")
ds

This came up at https://discourse.pangeo.io/t/making-kerchunk-as-simple-as-a-toggle/3690.

Read xarray's how-to-add-new-backend guide and the prototype built around xarray.backends.BackendEntrypoint. Start from the xarray.open_dataset(..., engine="kerchunk") entry point and determine how the kerchunk reference formats are recognized and opened. Done means the backend is usable through that API for the described references, with appropriate project coverage.