Do not use ignore flags in TNNMG
This commit partly reverts the last merge. We can't get ignore nodes in the tnnmg step, so we have to use 'classic' truncation by placing zeros in the matrix and rhs.
-
to avoid losing the recompute field in the MG transfer, we determine the ignore nodes out of the given zeros.
-
we carefully select zero diagonal entries in the UMFPACK base solver to circumvent singular matrices -> if the matrix is still singular, we try to regulize it.
-
we can't use direct Gauss-Seidel any more, therefore use TruncatedBlockGSStep.