Slow PFCOUNTs for usage data when collecting data for huge list of events

Summary

The PFCOUNT issued for collecting usage ping data tries to retrieve HLL counts for more than 1000 keys.

This is visible in the slowlog: https://log.gprd.gitlab.net/goto/fe18744f1fdd4334e105cc72a140409c

I think this might happen when we collect events for all categories: in Gitlab::UsageDataCounter::HllRedisCounter.unique_events_data. Specifically this seem to be caused by the issues_edit category.

Possible fixes

Maybe we should limit the number of keys we pass in one call, similar to how we do this for the PFADD