AMPI_Get_accumulate is broken
AMPI's implementation of MPI_Get_accumulate is completely wrong, and will always produce incorrect results at the target.
The control flow should look like:
1. Send the org buffer to the target (ie invoke a [sync] entry method, ampi::winRemoteGetAccumulate).
2. At the target, send back the contents of the targ buffer
3. At the target, accumulate the org buffer into the targ buffer
4. On the sender, recv back the original contents of the targ buffer, and copy them into the res buffer