Slow Diagonalize Supercell

[mesonepigreco]

DiagonalizeSupercell should be much faster for diagonalize phonons, with an expected linear scaling with the number of q points, however, the real timing scaling for very big number of q points is between N^2 and N^3, which is unacceptable and becomes the whole bottleneck of the calculation for SSCHA simulations of thousands of atoms.

Maybe that subroutine should be rewritten in julia, or more care must be taken to what is actually taking so long to execute.

With N = 1080 (6x6x6 q mesh)

Function: DiagonalizeSupercell
            N = 29 calls took: 0 hours; 1 minutes; 29.18 seconds
            Average of 3.075253026238803 s per call

With N = 320 (4x4x4 q mesh)

 Function: DiagonalizeSupercell
            N = 43 calls took: 0 hours; 0 minutes; 7.95 seconds
            Average of 0.1848654358886009 s per call

[mesonepigreco]

An improved version of DiagonalizeSupercell could only diagonalize the irreducible q points (thus enhancing a huge speedup), then retain all the remaining polarization vectors exploiting the symmetries.