In establish_cursors_test.cpp, InterruptedWithDanglingRemoteRequest has been occasionally hanging. It looks like a bug in the test: if establishCursors gets responses from both shards before it sees that its opCtx is killed, it's allowed to return successfully instead of throwing and cleaning up.

I think we need a second failpoint, to ensure the test marks the opCtx as killed before establishCursors does its check.