We need to expose a knob to disable the killCursor behavior in concurrency test suites because killCursor can be triggered by garbage collection at any time, which may be blocked by strong global locks that are waiting behind running transactions. Those running transaction may be running in the same thread as the garbage collection, e.g. in the shell, so none of them can make progress. When the transaction times out, the problem will go away, but since we increased the timeout in concurrency_simultaneous_replication to very large, the test will hang.