data: Yield the CPU in the NBX exchange loop under oversubscription

nbx_exchange polled with a tight 'while True: Iprobe' busy-wait, spinning
at 100% CPU while waiting for peer messages and the termination barrier.
When ranks are oversubscribed (more ranks than cores), a spinning rank
starves the very peer it is waiting on, so MPI makes no progress and the
exchange deadlocks. This is exactly the MPI-notebook CI configuration: a
4-engine ipyparallel cluster on a 2-core runner (OpenMPI --oversubscribe),
where it manifested as a multi-minute cell timeout.

Yield (time.sleep(0)) whenever a poll pass finds nothing ready, so
co-scheduled ranks can run; drain ready messages first via .
Verified: correctness unchanged (44 routed/gather mode-4 tests pass) and
16 ranks on 8 cores complete in 0.35s instead of hanging.

Author: mloubout

devito/data/distributed/transport.py

@@ -8,6 +8,8 @@
 graph communicator without affecting the layers above.
 """
 
+import time
+
 import numpy as np
 
 from devito.mpi import MPI
@@ -83,12 +85,19 @@ def nbx_exchange(comm, sendbufs, dtype, tag=0):
             buf = np.empty(count, dtype=dtype)
             comm.Recv([buf, mpitype], source=src, tag=tag)
             recvd[src] = buf
-        elif barrier is None:
+            # Drain any further ready messages before yielding
+            continue
+        if barrier is None:
             if MPI.Request.Testall(sends):
                 # All my sends were matched -> announce I am done sending
                 barrier = comm.Ibarrier()
         elif barrier.Test():
             # Everyone is done sending and nothing is in flight
             break
+        # Nothing was ready this pass: yield the CPU so co-scheduled ranks can
+        # make progress. Without this the probe loop busy-waits at 100%, which
+        # deadlocks the exchange when ranks are oversubscribed (more ranks than
+        # cores, e.g. the 4-engine ipyparallel cluster on a 2-core CI runner).
+        time.sleep(0)
 
     return recvd