pml/cm: pack data from application buffer in successive MPI_Start calls

This patch fixes a bug exposed by MTT MPI_Bsend_init_overtake_c, where
- MPI_Bsend_init(buf, ..., req)
- MPI_Start(req)
- MPI_Wait(req, stat)
- modify(buf)
- MPI_Start(req) // this does not send the updated data
- MPI_Wait(buf)

This is because PML/CM uses a bounce buffer send_req.req_buff to pack
data from  the application buffer in MPI_Bsend_int but not MPI_Start.

This patch adds a step to pack data from the application buffer in each
MPI_Start call.

Signed-off-by: Wenduo Wang <wenduwan@amazon.com>

Author: wenduwan

ompi/mca/pml/cm/pml_cm_sendreq.h

@@ -375,6 +375,10 @@ do {                                                                    \
             iov.iov_base = (IOVBASE_TYPE*)sendreq->req_buff;            \
             max_data = iov.iov_len = sendreq->req_count;                \
             iov_count = 1;                                              \
+            opal_convertor_prepare_for_send( &sendreq->req_send.req_base.req_convertor, \
+                                             sendreq->req_send.req_base.req_datatype,   \
+                                             max_data,                  \
+                                             sendreq->req_addr);        \
             opal_convertor_pack( &sendreq->req_send.req_base.req_convertor, \
                                  &iov,                                  \
                                  &iov_count,                            \