Prior to SERVER-106567, we would only call into the bucket_catalog::drop code for a collection if its namespace was a time-series buckets namespace. After SERVER-106567, we do this unconditionally. The reasoning behind this was that this would prevent us from looking up potentially stale information about the collection (i.e, it may have already been dropped and looking it up to check if it was a time-series collection may not have worked), and it was believed that there should be no side-effects to doing this more often.

However, performing this drop unconditionally for any dropped collection does have the side effect of always incrementing the bucket catalog era.

This grows the clearedSets vector more than before, but since we periodically clear this vector this shouldn't be a concern. We do bump the catalog era more often than we expect to, which prior to SERVER-108089 could cause us to fail reopen a bucket due to a catalog era mismatch error, ultimately allocating a new bucket instead, but after SERVER-108089 this shouldn't affect bucketing behavior.