Let's use otel bare metal runners for our benchmark codespeed tests. This way, we will be able to push more accurate results and move to a better position for comparing runs in the future.

Let's not use them for PR bench tests (150% comparison), as OTEL states:

Note: There is only one bare metal runner at this time, so please ensure your workloads are optimized to avoid unnecessary usage.

Interesting article from @kakkoyun with more improvements https://kakkoyun.me/posts/fosdem-2026-measuring-software-performance/