Adjust the criteria for the different errors. All the norms are equivalent the right constants are used.

Fix bug in the energy computation. Test works on MacOS X. But the results are strange. The force computations are more precise than the potential computation !!! We should investigate that.

# By Berenger Bramas
# Via Berenger Bramas
* 'master' of git+ssh://scm.gforge.inria.fr//gitroot/scalfmm/scalfmm:
  forgot to add private
  update starpu cuda
  specifies long long int as FSize

Rename utestChebyshevMultiRhs in utestInterpolationMultiRhs and test on Lagrange works but not Chebyschev one.

Update cuda starpu code by creating an accessor class (will certainly need an align on the cpu class)