Low-Rank restricted to the subspace of moment matrices

[blegat]

The low-rank matrix we get here:

MultivariateMoments.jl/src/extract.jl

Line 45 in 62e50f5

chol = low_rank_ldlt(M, dec, rank_check)

is maybe not a valid moment matrix.
If we project, e.g., by replacing the entries corresponding to the same moment by an average, we might not be low-rank anymore so we should integrate this with the LDLT probably.