Changes

Add GPU timing for MatILUFactorNumeric_SeqAIJCUSPARSE_ILU0. Reported-by: Hao Luo

Fix bug in PetscViewerBinaryWriteReadAll() that

prevented saving dense matrices whose number of entries on MPI rank 0 is greater than PETSC_MPI_INT_MAX by
removing unneeded PetscCall(PetscMPIIntCast(count, &cnt)); on rank 0.

Also removed unneeded potentially large malloc when used with on one MPI rank

Reported-by: Sreeram R Venkat <srvenkat@utexas.edu>

Code still does not work if any MPI process besides rank 0 has a matrix with a number of entries greater than PETSC_MPI_INT_MAX