@MichaelChirico can you review the PR and in particular think about more cases for tests that I might not have considered? Actually I first have to fix things...

@lorenzwalthert please ping again when it's ready, thanks!

Rebasing because the logs from the last workflow runs have expired.

thanks. My attempt was not working as expected and I don't have the bandwith to work on this at the moment.

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 8b2b788 is merged into main:

• 🚀cache_applying: 152ms -> 148ms [-5.73%, -0.6%]
• ✔️cache_recording: 513ms -> 513ms [-0.93%, +0.8%]
• ✔️without_cache: 981ms -> 982ms [-0.67%, +0.85%]

Further explanation regarding interpretation and methodology can be found in the documentation.