IT: allow out-of-order in kill_replica

[dorjesinpo]

kill replica can result in out-of-order in the following scenario:

0. consumer -> proxy -> replica1 -> primary.
1. kill replica1
2. proxy detects replica1 disconnect before primary does, and opens the queue on replica2
3. replica2 opens the queue on primary before primary detects replica1 disconnect, resulting in primary having 2 downstreams.
4. primary round robins messages {m1, m2}. m1 goes to replica1, m2 goes to replica2
5. primary detects replica1 disconnect and redelivers m1 to replica2 resulting in out-of-order {m2, m1}

[dorjesinpo]

assume IT failed because of missing sync point. New PR will enforce mqbnet::Channel flush on shutdown