Add eval seed timing logs and devservices mode

[JoshFerge]

Summary

• add an evals devservices mode for seeded local evals
• keep the mode focused on error-event eval ingestion: Relay, Snuba/Postgres, taskbroker, ingest-events, post-process-forwarder-errors, and dedicated taskworkers for relay, ingest.errors, ingest.errors.postprocess, and issues
• add seer_eval_seed.* timing logs for eval project ingest, save, post-process/code-mapping enqueue, polling endpoints, auto source-code config, and Relay project-config builds
• scope logs to eval-* projects where the code path has project context

Why

This is a draft/debug PR for the local seeded Seer eval bottleneck. The logs showed ingest-events consuming the 50 events within a few seconds, while sentry.tasks.store.save_event could start roughly 63s later because ingest.errors work was competing with issues tasks on the catch-all taskworker.

The new evals mode moves the queue topology into Sentry instead of requiring Seer to run manual taskworker commands. It avoids the broad ingest mode’s extra transactions/attachments workers and avoids a generic catch-all taskworker for the eval path.

Verification

• Sentry pre-commit hooks on commit and push
• git diff --check
• .venv/bin/prek run -q check-yaml --files devservices/config.yml
• devservices config loader check for evals mode and supervisor dependency types
• Used local logs to identify:
  • ingest consumer queue lag around 3.2-3.8s in the slow run
  • save_event since_received_seconds around 63s in the slow run
  • improved seed probe after dedicated workers: 11.93s, then 18.72s