fix(agent-monitoring): don't extrapolate span errors in traces table

[shellmayr]

• Had a look at some examples, and there are more than a few examples that show e.g. 64 errors for a trace with 1 LLM call, and when looking at it using this branch, it is a single error. This is not tenable of course, but this PR does introduce additional latency for large orgs, because it needs to scan all of the data

Closes TET-2082

[linear-code]

TET-2082 Agent monitoring dashboard - traces table extrapolates error counts