Optimize processing time of country calibration

The calibration of population and structural values per country is a computationally expensive task. Currently, one the function AbstractExposure.get_country_sum_asset_values is called which queries the sum of the population and the sum of the structural value of all assets of all entities in tiles of a given ISO code. This is a very expensive query due to the two INNER JOINs and may fail for very large countries due to memory problems. These values are used to compute the calibration factors. Then, the function AbstractExposure.get_country_entities is called to query all entity IDs in tiles of a given ISO code. These entities are then given to multiprocessing workers that calibrate the assets.

I suggest to rearrange the process as follows:

• Sum the population and structural value during the regular tile processing
  • After the buildings and the residual tile has been processed, a function can retrieve the sum of both values for all entities of the processed tile
  • Sum these over all tiles
• Calibrate all assets
  • Retrieve all Quadkeys/tiles
  • Start a multiprocessing run to calibrate all assets of all entities of a tile (each worker processes one tile)

\rfc @tara @laurens @kasra