Further investigate the dsend issue, in vain so far

Further investigate the dsend issue, in vain so far

- Ensure that it still compiles after the messed up merge (sorry)
- Add a simple dsend example
- Stop using the src_data to store the cleanup function since that
  field may be clear in MSG, but we store the MPI_status in there. So,
  add a cleanup_fun field down in the stacks so that both the cleanup
  function and the MPI_status can be stored
- Make that shit much more verbose (some messages should be removed
  once the bug is found, sorry for sharing my half backed changes)