Improve performance for a fast run

[facusapienza21]

@JordiBolibar if we want people to use this tool, we need to make it fast! I made changes in the organization of SphereUDE (similar to what we have in ODINN) and it would be great to make this code faster so we can run a good regression in a few minutes.

I think having custom Adjoints will definitively help, but I wonder where else in the code we can improve performance. Some notes I took from the SciML style documentation are

• Out-Of-Place and Immutability is preferred when sufficient performant
• Avoid globals variables
• Closures should be avoided whenever possible
• Avoid bounds check removal, and if done, add appropriate manual checks
• I am sure I am missing other things, but it would be nice to take a look at the code and see where we can get a significant boost in performance. Notice we have benchmark/run.jl to test runtime/memory use of the different AD methods we use.

[facusapienza21]

@JordiBolibar take a look to #149 to see the current runtime / memory usage of each regularization method. We need to make this faster!

[JordiBolibar]

I think the next important thing would be to debug type instabilities. This is pretty painful to do, as it requires time and a good understanding of the architecture. I have created a script to do so in my PR. Feel free to take a look, since I won't have much time to work on this in the coming days.