ensure_gradedness uses a full MPI_allreduce

[Thomas Engels]

The current version of ensure_gradedness uses a full MPI_ALLREDUCE:

call MPI_Allreduce(my_refine_change, refine_change, lgt_n, MPI_INTEGER1, MPI_MAX, WABBIT_COMM, ierr)

this can be extremely expensive if params%number_blocks becomes large and should be replaced b the more recent code using

subroutine synchronize_lgt_data( params, lgt_block, refinement_status_only )

[Thomas Engels]

closed for some time actually