Gradient-Based Optimization from cas_simplification branch

[nightdr]

Includes changes involving the implementation of gradient-based optimization methods for continuous local optimization. I started the architecture for this optimization and implemented it in the test cases and with explicit regression.

[nightdr]

I added changes to fix up the issues brought up in the initial review.

In terms of other issues that we discussed in this week's meeting:

• I believe the culprit of the "BFGS" and "CG" algorithms with the absolute value case was the gradient tolerance needed for each method (i.e. the infinity norm of the gradients need to fall below a certain threshold, defaulted to 1e-05, for convergence, see https://docs.scipy.org/doc/scipy/reference/optimize.minimize-bfgs.html and https://docs.scipy.org/doc/scipy/reference/optimize.minimize-cg.html for more details). I changed the respective integration and unit test cases to work with functions that have decreasing gradients, but functions with constant gradients might cause issues in the future when using these algorithms.
• In regards to using FitnessFunction properties in the GradientMixin classes, it seems that it is convention to assume that mixins will be paired with some classes and be able to use those classes' properties. Furthermore, I did not change get_jacobian() to return both the fitness vector and the jacobian because it would complicate using get_jacobian() in the scipy root call for continuous local optimization. However, I could add a wrapper function for the scipy root call if you think that it would be beneficial to return both the fitness vector and jacobian in the function.

[nightdr]

I made the changes we discussed during our meeting. However, I left the inheritance order as mixin classes first then the base class. I left the order like this because there is no super call in FitnessFunction I could do to pass a metric to a mixin (i.e. If we inherit the base class then mixin classes, there has to be some super call that passes the parameters from the base class to the mixins. However, this doesn't seem possible without adding a metric parameter to the FitnessFunction constructor). It also seems that in other libraries like Django, mixins are inherited first followed by the base class. However, let me know if you want me to try to figure out how to inherit the base class first then the mixins.