Add evaluation timings

[frostedoyster]

Adds timings (per atom) to the evaluation script, as in #338

Contributor (creator of pull-request) checklist

• Tests updated (for new features and bugfixes)?
• Documentation updated (for new features)?
• Issue referenced (for PRs that solve an issue)?

---

📚 Documentation preview 📚: https://metatrain--353.org.readthedocs.build/en/353/

[frostedoyster]

@spozdn as well (can't request your review for now)

[abmazitov]

Thank you @frostedoyster , this is indeed very useful!!

I have two questions: shall we also add the std value of timing to check if the performance is uniform? And do we have the evaluation done with the batch size of one, or we can actually set the batch size?

[frostedoyster]

I like adding the std (although one should keep in mind that it might be high if the sizes of the structures aren't uniform). We're still using a batch size of one, and this is still hardcoded (the user can't change it).

Adding a batch size for eval would be a bit challenging for our API because eval, for now, takes no options. This means that we should have options for eval and, since eval is called after train, an eval section in the training options. I think that should be part of a different PR. I think this PR is still useful however because one structure at a time is pretty much what you would get if you did MD on the test set.