Aggregator enforcing project limits by partition may be too harsh

[Dav1dde]

A per project limit is now enforced only per partition, this assumes an equal distribution of all buckets across all partitions, but this seems currently to be too strict.

We're seeing limits being enforced with the total size of the aggregator being way below the maximum, which hints towards an uneven distribution of buckets for projects in partitions. An example for this could be a very high volume span produces a lot of expensive duration metrics which all get aggregated into a single partition.