This counter counting usage by project, resulting in an ever growing
set in Redis as the feature gains adoption.

This data wasn't useful, while still consuming a relatively high
amount of memory, so we removed it.

A better approach would be using Redis HLL counters for this.