CRS handling API

[benbovy]

A tentative API for all CRS specific operations:

• Dataset.proj.assign_crs(...) : (re-)set a CRS without data transformation
  • create the spatial reference coordinate if it doesn't exist yet
  • see Method name for setting a CRS #3
• Dataset.proj.guess_crs(...) : try setting a new CRS from the metadata of an existing spatial reference (scalar) coordinate.
• Dataset.proj.to_crs(...) : set a new CRS with data transformation (re-projection)
• Dataset.proj.write_crs_info(...) : add metadata to a spatial reference coordinate given some "flavor"
  • e.g., CF grid mapping attributes, crs_wkt, etc.
  • Accessor: add methods to write or clear CRS info #20
• Dataset.proj.clear_crs_info(...) : remove metadata from a spatial reference coordinate
  • Accessor: add methods to write or clear CRS info #20

Alternatively we could expose the same functionality through fewer methods, e.g., assign_crs would guess the CRS from coordinate metadata if no CRS is given, assign_crs may also write metadata depending on a keyword argument, etc. I think I prefer the suggestion above, though (more methods with limited scope and clean API).

We should probably avoid being too opinionated here about how CRS is represented in metadata, but since pyproj supports many formats it would make sense for xproj (= Pyproj Xarray wrapper) to provide convenient API like guess_crs, write_crs_info and clear_crs_info.

[benbovy]

Dataset.proj.to_crs(...) may be tricky to support here since it usually involves operations such as resampling (regridding) on raster (gridded) data, which are applied on data variables and may be done in a lot of different ways. It is also not clear to me how it would work together with Dataset.proj.map_crs(transform=True) added in #19.

It is easier if we first restrict XProj's scope to handling spatial reference coordinates and possibly other CRS-aware indexed coordinates via Dataset.proj.map_crs().