ref(spans): Track cumulative latency in buffer logger

[fpacifici]

Changes BufferLogger to track cumulative latency instead of maximum latency, providing better visibility into which project/trace combinations accumulate the most latency over time.

I think tracking the cumulative time spent by outlier traces is going to be more valuable than specific traces that take longer in specific instances of the EVALSHA call.
This will allow us to keep constant track of outlier traces so we can take action wether manually (killswitch) or automatically (demote the trace to BLQ).

The idea is to keep track of the top traces and print the top 50 periodically.
Then identifying outliers will become easy.

Changes

• Remove threshold filtering: All entries are now tracked regardless of latency value
• Track cumulative latency: Store sum of all latencies for each project_and_trace instead of just the maximum value
• Smart trimming: After each batch, if entries exceed 50, trim to keep only the top 50 by cumulative latency (removes lowest latency entries)
• Sort by cumulative latency: Periodic logging (every 1 minute) now reports top 50 entries sorted by cumulative latency

Made with Cursor

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}