Fetching Prometheus metrics is slow and CPU-intensive

## Problem

Profiling shows that the Prometheus metrics scrape endpoint is unexpectedly slow and consumes significant CPU — visibly impacting application performance during scrape intervals.

DataDog APM flamegraph shows the metrics collection path as a major hotspot. Needs investigation to identify whether the issue is in metric aggregation, serialization, or the underlying data queries.

🤖 Filed by Claude on behalf of Kris.