Reduce latency for metric ingestion

[jan-auer]

The ingestion pipeline causes a delay of 1-2 minutes for metrics. Users expect metrics to show up earlier in the UI.

Especially if you want to have alerts or reactive functionality like canary deploys based on DDM, metrics need to be as high resolution and as low latency as possible. Since DDM has a resolution of 10s resulting from its bucketing window, users expect an ingestion delay roughly on the same level.

The delay results from bucketing and additional delays in both PoP and processing Relays. The theoretical minimal delay with current configuration is 70s in addition to any delays that the indexer and snuba consumers add.