Query initiated at slave MDB are stuck

The solution for MCOL-5279 was not good enough to resolve the MCOL-5279 for the queries started on MDB slaves in a cluster.
The scenario mentioned in MCOL-5279 doesn't work if run at slave MDB.
The MCOL-5279 solution made an assumption that the primary will have a fixed index always that is no the case when queries run at MDB slave nodes in a cluster.
More generic approach makes an assumption there will be only one local connection b/w ExeMgr and PrimProc. DEC methods that use DEC::writeToClient searches for this local connection index and sends the message into this local connection in the last turn. This is possible b/c DEC:: methods synchronously send messages to cluster nodes.