fix: Throughput processor should be singleton

[BinaryFissionGames]

Proposed Change

• Make individual throughput processors a singleton across multiple pipelines.
• If a throughput measurement processor is registered twice, throw an error.
• Fix issue where empty metrics were emitted for every 0 valued metric.
• Don't log every time we send metrics.

We were seeing some issues with measurements in BPOP. This was because when the same processor was included in multiple pipelines, it would create multiple instances. In that case, one processor would "win" the registration, and only measurements for that particular pipeline would be emitted.

This change makes it so that there is only one instance of the processor (this pattern is copied from the snapshot processor), so all the pipelines are accounted for. You can see that here in BPOP, where we have multiple sources (which end up being multiple pipelines) adding up correctly.

Checklist

• Changes are tested
• CI has passed

[antonblock]

Working well with the custom message change enabled in BP, just had a couple small comments