adds adjoint to metalens notebook

[tylerflex]

This adds a new section to the visible light metalens where we use local field projection and adjoint to fine tune the power directed to the focal spot.

We get a 17% improvement over 15 iterations.

NOTES:

• uses yaugenst-flex/polyslab-grads frontend branch, which has not been merged. without the code from this branch, the results will be terrible because the gradients are inaccurate.
• uses local gradient calculation to make use of the changes in this branch. Ideally we would switch to remote gradient.
• The gradient processing takes >5 minutes each iteration.. so I limited the code to only run one iteration. I left the cell output to correspond to the full 15 iterations that I ran locally, along with a small note.

Not sure when is the best time to merge this, maybe after Yannick's PR is merged and it is fast enough to justify running this?