Now optimizing currents, using diffrax, optimizing with least_squares and adding current penalty

[rogeriojorge]

This optimization with least-squares appears to be able to optimize particles from a cold start, starting from a tokamak to a stellarator.
The loss function includes a current penalty so that now that currents are optimized, they do not become too big.

[rogeriojorge]

The following changes were made:

• There is an overall factor of 7e6 in the magnetic field. This allows us to change the current dofs using order 1 numbers instead of order 7e6
• Currents are now being optimized. However, not the first coil current as these can increase together, leading to a larger magnetic field overall and better confinement that way
• The tracing is done mainly on trapped particles, as the range of v_parallel was restricted only between -0.2 and 0.2 times v_thermal.
• Least squares optimization appears to be stable, yielding better results than Adam. However, for now, both are being used in each optimization
• Coils start to get increasingly bigger with optimization, regularization penalties may be needed.

[EstevaoMGomes]

What are the comments in functions "BdotGradTheta", "BdotGradPhi" and "BcrossGradBdotGradTheta" for?

[EstevaoMGomes]

Does not seem to improve the efficiency.
Tested for an averaging of 10000 particles with the "compare_simsopt" file.