PCBDDC: fix for getri error -2 with MKL

MKL seems to check the pointer (NULL) for the matrix even when lwork == -1

PCBDDC: do MatMultTranspose only if needed

Move PCBDDCSubsetNumbering to ISRenumber

This function is needed by MatGetLocalSubMatrix_IS

PCBDDC: use IS instead of ISLocalToGlobalMapping

PCBDDC: Added API for passing the matrix representation of the divergence operator

PCBDDC: fixes for sequential runs and zero coarse size

PCBDDC: add extra dbg output

Merge branch 'stefano_zampini/feature-fetidp-class'

PCBDDC: propagate divudotp to coarser levels

Tested with mixed boundary conditions with 3 additional levels

PCBDDC: forgot

PCBDDC: add internal API to compute no-net-flux condition

Merge branch 'stefano_zampini/feature-pcbddc-saddlepoint' of https://bitbucket.org/petsc/petsc into stefano_zampini/feature-pcbddc-saddlepoint

PCBDDC: prevent from writing on standard_rhs that could be locked

PCBDDC: bugfix

PCBDDC: no need to create field split with a single field

PCBDDC: fix set but not used warning

PCBDDC: create sub_schurs just when needed

PCBDDC: don't modify mat_ctx->temp_solution_D

KSPFETIDP: allow customization of multipliers from KSPSetFromOptions

KSPFETIDP: initial implementation of FETI-DP classes

FETIDP implemented as a Krylov solver, since it solves a different linear system (innerksp)

FETIDP: check number of multipliers with PETSC_USE_DEBUG

FETIDP: added lumped preconditioner

FETIDP: fix for benign subspace trick. Now FETI-DP works with the benign subspace trick

PCBDDC: fix memory leak

FETIDP: fetidp mat mult should zero p0