AsyncResultMerger should not retry killed operations

As part of the changes to make sharding more robust in the presence of CSRS config server stepdowns we made most operations retriable on these error codes.

This is problematic because we have no way of differentiating between interruptions due to the actual operation being killed or due to the primary stepping down, so now operations fail to get killed as experienced by failures in jstests/sharding/features3.js.