Speed up cubes.py

[cgueret]

Instead of loading the output in RDFLib just store it directly on disk (we get a turtle reply from the triple store) and issue a query at the end of the process to see which dimensions are used by the data points