ratelimits: Remove a metric and some labels that we're not finding useful

[beautifulentropy]

Fixes #7879

[jsha]

Nice cleanups!

We had talked in standup specifically about removing the batchset_entry, batchsetnotexisting_entry, and other observations that try to backformulate the per-request latency based on overall pipeline latency (which has some potential to be misleading since the costs don't divide out evenly). I see this PR goes one step farther and removes the whole ratelimits_latency metric.

I'm okay with that; we also have the ratelimits_spend_latency metric:

boulder/ratelimits/limiter.go

Line 49 in c99b657

Help: fmt.Sprintf("Latency of ratelimit checks labeled by limit=[name] and decision=[%s|%s], in seconds", Allowed, Denied),

. That one is different in that it can span multiple calls (Get, BatchSetNotExisting, etc). But on balance it probably suffices. I see that the Redis client also has otel instrumentation which may be a nicer way to get per-Pipeline() latencies anyhow, once we start using otel internally.

[beautifulentropy]

We had talked in standup specifically about removing the batchset_entry, batchsetnotexisting_entry, and other observations that try to backformulate the per-request latency based on overall pipeline latency (which has some potential to be misleading since the costs don't divide out evenly). I see this PR goes one step farther and removes the whole ratelimits_latency metric.

Oh, thanks for bringing this up! I actually thought that discussion started with a proposal to remove this metric. Removing just per_entry makes a whole lot of sense, I'm quite happy to revert most of these removals. Let me throw something together real quick.