Extend the ESM collection specification to support catalogs containing datasets in GRIB format

**Is your feature request related to a problem? Please describe.**

Currently, the only valid data formats supported by intake-esm are `netcdf` and `zarr`. As a result, it's not possible to use intake-esm when working with datasets in grib format.

**Describe the solution you'd like**

We should consider extending the esm collection specification: https://github.com/intake/intake-esm/blob/main/docs/source/explanation/esm-collection-spec.md#assets-object to allow catalogs to contain datasets in grib format.

**Additional context**

I'm copying and pasting the comment in https://github.com/intake/intake-esm/issues/66#issuecomment-1061610199

I made one for dkrz era5 data which is on our HPC's disk storage. That means, the data access also only works on Mistral.

```
dkrz_cdp=intake.open_catalog("https://swift.dkrz.de/v1/dkrz_a44962e3ba914c309a7421573a6949a6/intake-esm/dkrz_data-pool_cloudcatalog.yaml")
esm_dkrz_era=dkrz_cdp.dkrz_era5_disk_grb_fromcloud
```
should work.

However, it is based on raw `grb` data and `intake-esm` does not really support `grb`. We would need a default engine [here](https://github.com/intake/intake-esm/blob/40fe3a7808c14af77d96cfc932b9cebaa6beba10/intake_esm/source.py#L18) for grib. The one i use is `cfgrib`. You may say that users can provide that but since there is no `grib_kwargs`, I cannot for example merge `grb` assets with `netcdf` assets. Which is why I say there is no `grb` support in intake-esm.

Best,
Fabi

_Originally posted by @wachsylon in https://github.com/intake/intake-esm/issues/66#issuecomment-1061610199_

Read the assets object section of docs/source/explanation/esm-collection-spec.md and inspect intake_esm/source.py, especially the engine handling referenced in the issue. Determine how GRIB catalogs and grib_kwargs should be represented, including mixed GRIB and NetCDF assets. Done means the specification and loading behavior support GRIB datasets consistently.

Written by the indexing model from the issue text.