Use Gram-Schmidt for create_mrt_orthogonal

• orthogonalize automatically if no moments from literature are provided
• must specify whether weighted or unweighted orthogonality is desired
• add lbmpy.moments.is_bulk_mode and make lbmpy.moments.is_shear_mode also detect the diagonal modes
• use the bulk mode (x^2+y^2+z^2-1) in orthogonalization so we get the same physical modes as in literature

We now match the literature MRTs in all the physical moments. The only exception is D3Q27, which chooses a different linear combination for the shear moments. Differences in the higher-order moments remain, but they have no effect on the model.

Fixes #5 (closed)