Neural Network-based cable spline avoidance

[kshaji3]

Key additions:

1. Adds in a method for using neural network based dynamics prediction into GT Dynamics.
2. Adds a factor that allows you to have a neural network-based model of cable spline prediction (as a psuedospectral chebyshev polynomial) attached to the robot body as a collision constraint.
3. Workshop paper was submitted on this framework.

[dellaert]

Requesting changes based on correctness, performance, API design, and GTSAM style/test-organization guidance. The relevant C++ tests and all six CI jobs pass, but the inline issues below should be addressed before merging.

[kshaji3]

Finally passed CI!