Add support for geostationary projection datasets (Himawari-8 example)

[eeholmes]

Dataset

https://data.source.coop/bkr/geo/himawari_1000m.icechunk

Problem

Gridlook currently supports several grid/projection types (regular, curvilinear, polar_stereographic, healpix, triangular, etc.), but does not yet support geostationary satellite projections.

This dataset is a Himawari-8 AHI full-disk geostationary dataset stored in Icechunk/Zarr. The dataset contains projected x/y coordinates in meters and enough metadata to reconstruct the projection. BUT the xy coords are called x_geostationary and y_geostationary.

Relevant metadata:

{
  "proj": "geos",
  "lon_0": 140.7,
  "h": 35785863,
  "x_0": 0,
  "y_0": 0,
  "a": 6378137,
  "rf": 298.257024882273
}

Coordinate ranges:

x_geostationary: -5.5e6 to 5.5e6 meters
y_geostationary: -5.5e6 to 5.5e6 meters

Proposed catalog entry

{
  "title": "Himawari-8 AHI Full Disk Reflectance",
  "url": "https://data.source.coop/bkr/geo/himawari_1000m.icechunk",
  "format": "Zarr v3",
  "access": "icechunk",
  "convention": null,
  "layout": "simple",
  "grid": "geostationary",
  "crs": "geos"
}

Notes

The dataset uses extremely large chunks:

chunks: (1, 11000, 11000)
dtype: float16

This corresponds to approximately:

~242 MB uncompressed per chunk

which may make visualization slow / impossible with gridlook.

[eeholmes]

prob similar problem https://data.source.coop/bkr/nsrdb/nsrdb_india.icechunk