Supporting the new USA model

[micheles]

There is still work to do:

• reading the sections.xml file is slow
• Estimating weights for the atomic groups is slow, we should reduce the sitecol
• We should have the option to run one specific atomic group
• Can we have atomic groups not using the colon convention?
• Why so much memory is needed? It is due to the multifaultsources
• Fix the OverflowError: Python integer 4679680567 out of bounds for uint32
• The cluster loop in the PmapMaker is pretty slow and memory consuming

NB: there is nothing to do about the first point since it is due to KiteSurface.from_profiles and not to XML parsing, except
using a large enough rupture_mesh_spacing (i.e. not 1 km as it is now).