cross-backend processing: expand output format aspect

[soxofaan]

Spin-off/follow-up issue for #150 ("deep" graph splitting):

The current implementation only works if split is done so that intermediate result is a raster data cube, stored in GeoTIFF.

Expand this output format aspect. For example, some ideas to work on:

• also allow netCDF, also see crossbackend feature in aggregator: use NetCDF iso GTiff? #145
• also support vector cubes (e.g. to split after a aggregate_spatial)
• take back-end output format capabilities into account (instead of blindly picking GeoTIFF)
• let the user pick the preferred/allowed formats?