GEMMT default, bugfix with MKL
- Use GEMMT by default for symmetrix matrix-matrix multiplication, if available (e.g. Intel MKL)
- For 64-bit with MKL (MKL64), link "mpi_ilp64" as well in FindScalapack
- MKL with OpenMP: correct bug that stopped it from building.
- Cray: don't set compiler tune options, regardless of compiler