Add hwloc area by performance criterion
This PR enhances hwloc areas with a new aml_area_hwloc_preferred area. This area will hold a list of numanodes sorted by performance criterion. mmap operation will use first memory to map data block per block. When mapping fails (because the node is full), consecutive numanodes will be used to map data.
This only adds code.
- hwloc area is modified.
- Some utils about hwloc are added as well but meant for library internal use only. These utils will build a distance matrix based on topology hops as a fallback for when machines do not support performance criteria.
- Only the tutorial is used as a test case to save time on otherwise complex test scenario.