PSYNC2: discard pending transactions from cached master.

During the review of the fix for #3899, @yangsiran identified an
implementation bug: given that the offset is now relative to the applied
part of the replication log, when we cache a master, the successive
PSYNC2 request will be made in order to *include* the transaction that
was not completely processed. This means that we need to discard any
pending transaction from our replication buffer: it will be re-executed.

src/replication.c

@@ -2120,10 +2120,12 @@ void replicationCacheMaster(client *c) {
     unlinkClient(c);
 
     /* Fix the master specific fields: we want to discard to non processed
-     * query buffers and non processed offsets. */
+     * query buffers and non processed offsets, including pending
+     * transactions. */
     sdsclear(server.master->querybuf);
     sdsclear(server.master->pending_querybuf);
     server.master->read_reploff = server.master->reploff;
+    if (c->flags & CLIENT_MULTI) discardTransaction(c);
 
     /* Save the master. Server.master will be set to null later by
      * replicationHandleMasterDisconnection(). */