fix: apply log level/format before NewRunner() emits startup logs

[ucpr]

Summary

When LOG_FORMAT=json was set, startup logs emitted before the logger configuration was applied used logrus’s default text formatter.

This affected Stats-related logs emitted from NewRunner() and tracing initialization logs emitted early in Run(). Logs emitted later, such as the Redis connection log, used the configured JSON formatter. As a result, a single process produced logs in inconsistent formats.

time="2026-07-16T03:20:52Z" level=info msg="Stats initialized for Prometheus"
time="2026-07-16T03:20:52Z" level=info msg="Starting prometheus sink on :9090/rate-limit-metrics"
time="2026-07-16T03:20:52Z" level=info msg="Stats flush interval: 30s"
time="2026-07-16T03:20:52Z" level=info msg="TracerProvider initialized with following parameters: ..."
{"@message":"connecting to redis on 127.0.0.1:6379 with pool size 64","@timestamp":"...","level":"warning"}

Fix

Extracted the logger level/format configuration into configureLogger() and call it at the top of NewRunner(), so it's applied before any log line is emitted, including those inside NewRunner() itself. Removed the now-duplicated configuration code from Run().

Note: #428 was a prior attempt at the same root cause, but it only reordered the TracerProvider initialized log inside Run() and didn't address the Stats-related logs emitted from NewRunner(). It went unreviewed and was stale-closed. This PR covers both.

Test plan

• go build ./...
• go test ./src/...
• Ran locally with LOG_FORMAT=json and confirmed all startup logs (Stats initialized / Stats flush interval / Tracing disabled / etc.) are now emitted as JSON