Basically, fakelag was counting the time imposed by its own sleeps as though
the user had themselves paused for that amount of time. Therefore, if a user
sent a large number of consecutive commands, every other command would pause
for the expected throttle interval, but the subsequent command would be
processed instantly (you'd get two back-to-back commands). This resulted in
throttled users being able to send at double the expected rate.

* Placate `go vet`
* Reorder the `send` loop, clarify things a little

1. Implement persistent channel keys (#208)
2. Persist changes to registered channel modes
3. Backend changes for persisting operator status (partial #198)

The sync.atomic documentation says:

"On both ARM and x86-32, it is the caller's responsibility to arrange
for 64-bit alignment of 64-bit words accessed atomically."

See discussion on issue; this might not be a complete fix but it's
definitely better than the current behavior.