Adding gradient descent information to EngiBench problems

[gapaza]

For adding the recording of gradient descent information into each EngiBench problem, I wanted to propose a means for this that consistent across all problems. Specifically, I was thinking that we could add an addition field to the OptiStep data class (specifically a npt.NDArray), that contains a gradient descent step for an optimization routine. I would argue that this is an appropriate place, as the optimize function of all problems is designed (at least in the parent Problem class) to return the final design and a list of OptiStep objects already. Furthermore, the list of OptiStep objects is defined as containing the optimization history, which I think fits the bill. Of course, if there is a better way to do this I am happy to implement it as well. Once the format is agreed upon, my plan is to create a new branch for each problem, add the functionality, and submit a pull request.

[linear]

SOH-74

[markfuge]

As discussed in the meeting the plan is to include it in OptiStep as suggested, such that each step will now include x, f(x), and grad(f(x)). This may involve one extra gradient calculation at the end of a run (e.g., if it has converged already and a new gradient step wasn't needed), but we will prototype with what Gabe suggested and work from there to refine after some testing.