Changes to improve performance

[JordiBolibar]

PR for #150.

Changes implemented so far:

• Making the ODEProblem local instead of global
• Specifying a bit argument types.

Things I have tried:

• Adding parametric types to forward.jl functions. Since ReverseDiff uses TrackedArrays and values it just breaks everything. The abstract types are the most specific thing I managed to narrow down.
• Using EnzymeVJP(): this gave an ugly error which was not super straightforward to debug. Once we gain more experience with ODINN we can try to tackle this.

@facusapienza21 the benchmark has failed. In local it seems to work, so I'm not sure what failed.

We can work on this iteratively.

[facusapienza21]

Hi @JordiBolibar ! Thank you for taking a look at this! definitively some improvement in the code!

I am happy to see what break, but a few points:

• Notice that the repo style now follows the Julia standards and you are now breaking the quality check format. This is not critical for exploration, but I will prefer not to merge into main something that does not pass the check.
• Can you make a PR to the branch upPerformance instead? I created this branch to follow up on this.
• I now agree with what Alban say that is better to work in branch on the same repo that having the PR from the fork: I could directly jump in into your branch and make the changes to fix the PR, but since this is coming from your fork I cannot directly commit there.

[facusapienza21]

Just some small comments on the code, other than the comments I previously posted.