changed mpi_test to mpi_wait in forwardMessageSimple functions

[jjepson2]

Here I changed the mpi_tests's to mpi_wait's, to prevent the send buffer from being overwritten until it has been sent out.

[liuyangzhuan]

I didn't observe any significant slowdown in the triangular solve using this fix. But in case the code hangs in future, we can consider using separate buffers per tree.