Daskify geometry operations

[forman]

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

The xcube function xcube.core.geom.rasterize_features() seems to eagerly load large numpy arrays into memory often resulting in an OOM error.

xcube.core.geom.get_geometry_mask() may be affected too as it is effectively a wrapper for rasterio.features.geometry_mask() which may not use Dask at all (to be verified!).

Describe the solution you'd like

Instead they should produce lazy dask arrays.

[forman]

...which may not use Dask at all (to be verified!).

Verified, it does not use Dask at all and returns a numpy array. From rasterio docstring:

This function requires significant memory resources. The shapes
iterator will be materialized to a Python list and another C copy of
that list will be made. The `out` array will be copied and
additional temporary raster memory equal to 2x the smaller of `out`
data or GDAL's max cache size (controlled by GDAL_CACHEMAX, default
is 5% of the computer's physical memory) is required.

[forman]

Issue has been party addressed in #594.