Eigendecompose quindiagonal model for beta circular ensembles

[guilgautier]

Check eig_banded for quindiagonal model for beta ensemble with uniform measure on unit circle. Clue 0.5(L.dot(M)+M.dot(L))

[guilgautier]

eig_banded requires hermitian symmetry. This is neither the case for L.dot(M) nor 0.5(L.dot(M)+M.dot(L)).

In https://github.com/guilgautier/DPPy/commit/537b26dd90075c7e7be82d63565a8466ce2395d1 I simply used sparse representations of L and M for which:

• a block_diag method already exists, that's why I remove the hand-coded one
• dot method is more efficient for large size

However for a sparse array it is not possible to compute all its eigenvalues (see karg in eigs) that's why toarray() is used for now ☹️