[Bugfix] GDA/bnxt: release SQ lock before return (#372)

* bnxt_post_wqe_amo_single with fetching = true would return
  before releasing the send queue lock, resulting in a deadlock.
* Release the send queue lock before returning from the function.

src/gda/bnxt/queue_pair_bnxt.cpp

@@ -434,13 +434,13 @@ __device__ uint64_t QueuePair::bnxt_post_wqe_amo_single(uintptr_t raddr, uint8_t
 
   bnxt_ring_doorbell(sq.tail);
 
+  release_lock(&sq.lock);
+
   if (fetching) {
     quiet();
     return fetching_atomic[atomic_idx];
   }
 
-  release_lock(&sq.lock);
-
   return 0;
 }