subset_spatial op very slow with non-trivial polygons

[forman]

Problem

subset_spatial op is very slow (too slow) with mask=True and region being a non-trivial polygons and when many grid cells have to be compared.

This is probably because we compare for each grid cell individually if it intersects the polygon. Would be more efficient if we create the mask by rendering the polygon as a masked numpy array (binary image).

Steps

In Cate Desktop:

1. Select operation read_necdf and add step with with file name e.g. "20100401120000-ESACCI-L4_GHRSST-SSTdepth-OSTIA-GLOB_LT-v02.0-fv01.1.nc"
2. From countries layer select Germany (yes, SST is an ocean product, but still there are many coastal cells this way)
3. Select operation subset_spatial and add step with ds_1, the selected POLYGON and mask on.

This actually computes forever.

Specifications

Cate 1.1 dev1

[JanisGailis]

Note to self - start with profiling the code