Draft: better Slow/SlowOk alerts

instead of checking the last alert and only emitting SlowOk if a Slow was emitted, the root cause of the issue was pinned down: Slow alerts were being missed due to the transition from Ok to Slow was not caught.

the new algorithm works as such: Every slow_threshold rows, calculate the average latencies of one half and then the other half, and then calculate the transitions from the averages, instead of "is this entire half slow" vs "is this entire half not slow" like we did before.

this has been deployed to status dot lavatech dot top to validate this new method works. if it works in the next couple of days, this shall be merged.

close #44.