Avoid torchscript errors when float options are supplied as integers

[frostedoyster]

Closes #414. Implements a fix for the case where float options are instead supplied as integers

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--419.org.readthedocs.build/en/419/

[frostedoyster]

@abmazitov I need your help for PET, because I don't know where the best place to convert the parameters to float is. We could do it here (although it's a bit of a hack) or in the PET repository, as you wish

[Luthaf]

Is this something we could do using the json schema validation? I think there is a way to specify that a value should be a float and not an integer there, and we could maybe use this information to ask the validator to accept both and then do the conversion ourself.

This mean we would not have to do this (and think about it) for each new architecture.

[frostedoyster]

There is a way to specify that a number should be a float, but then the validation will fail, right?
I implemented the fix as you and others agreed in the issue

[Luthaf]

I was thinking of something where (a) we declare that we want a float in the schema, but (b) we modify the schema before validation to also allow for integers and (c) we do the conversion after the validation. This way the user does not have to think about it, and the architecture developers do not have to think about it. But it might be a lot of work for minimal gains, so the current fix should also be fine!