Fix replication queue full-detection to actually detect fullness

BoundedChannelFullMode.DropWrite makes TryWrite report success (and
silently discard the item) even when the channel is full, so the
"queue full -> report member unavailable" branch in Replicate() never
fired and the dropped barrier slot was never resolved, leaking that
ReplicationBarrier instance out of the pool. Wait gives TryWrite the
false-when-full return value the code already assumed.

Author: sakno

src/cluster/DotNext.Net.Cluster/Net/Cluster/Consensus/Raft/ReplicationUtils/ReplicationProcess.cs

@@ -44,7 +44,10 @@ public ReplicationProcess(TMember member, int queueSize)
 
         var channel = Channel.CreateBounded<ReplicationBarrier>(new BoundedChannelOptions(queueSize)
         {
-            FullMode = BoundedChannelFullMode.DropWrite,
+            // DropWrite makes TryWrite report success (silently discarding the item) even when full;
+            // Replicate() below relies on a false return to synchronously report the member as
+            // unavailable for this round, which only Wait actually provides.
+            FullMode = BoundedChannelFullMode.Wait,
             AllowSynchronousContinuations = false,
             SingleReader = true,
             SingleWriter = true,