We currently use the replication coordinator method waitForReplState only in the test command waitForReplState. It appears the library calling this uses assert_soon, thus it tolerates interrupts, but in fact this method is not interruptible. Since it's called with an opCtx active, that means it can block anything which requires opCtxs to be killed; currently the only example is storage change on initial sync.

We should make it take an Interruptible or OperationContext.