Adding stepwise regression feature to `PolynomialPathFitter`

[nickbianco]

Fixes issue #3761

[perf-win]

Brief summary of changes

Adds the option to use stepwise regression when fitting polynomial coefficients in PolynomialPathFitter. This approach fits polynomials using individual terms one at a time and keeps the terms that result in the lowest fitting error. Terms (and coefficients) are added until the path length and moment arm tolerances are achieved.

One key change was made to MultivariatePolynomialFunction: coefficients equal to zero are now ignored when evaluating polynomial functions to maximize speed.

Testing I've completed

• Ran Matlab and Python scripting examples, examplePolynomialPathFitter.m/py.
• Updated the FunctionBasedPath set used in exampleMocoInverse and exampleMocoTrack and ran both examples.

Looking for feedback on...

CHANGELOG.md (choose one)

• updated.

Performance analysis

Platform: Windows, self-hosted runner

Test Name	lhs [secs]	stderr [secs]	rhs [secs]	stderr [secs]	Speedup
Arm26	0.35	0.00	0.36	0.00	0.97
ellipsoid_wrap	4.12	0.00	4.15	0.01	0.99
ellipsoid_wrap_function_based_paths	3.41	0.00	3.42	0.00	1.00
Gait2354	0.42	0.00	0.42	0.00	1.01
MocoSlidingMass	1.48	0.00	1.48	0.00	1.00
passive_dynamic	5.61	0.00	5.59	0.00	1.00
passive_dynamic_noanalysis	3.45	0.01	3.41	0.00	1.01
RajagopalModel	8.89	0.01	8.88	0.01	1.00
ToyDropLanding	13.38	0.01	13.41	0.00	1.00
ToyDropLanding_fbp_stepwisereg	13.91	0.07	12.31	0.01	1.13
ToyDropLanding_function_based_paths	12.43	0.01	12.51	0.03	0.99
ToyDropLanding_nomuscles	0.57	0.00	0.56	0.00	1.01

---

This change is 

[nickbianco]

Thanks @pepbos! I made some updates based on your comments. Unless you have any other comments, I'll merge after the tests pass.