Improvements to `PolynomialPathFitter`

[nickbianco]

After recently fixing some issues in PolynomialPathFitter, it is becoming increasingly clear that the current implementation of the path fitting pipeline is a bit unwieldy and somewhat hard to maintain. Additionally, there are certain potential improvements to path fitting that the current implementation will make it hard to integrate. Below is a list of maintenance changes that would bring PolynomialPathFitter into a healthier state. Following that is a list of features suggested previously be Ton van den Bogert and other users.

Maintenance improvements

• Move private helper functions that are not heavily dependent on properties to an anonymous namespace in PolynomialPathFitter.cpp. (Refactor PolynomialPathFitter #4302)
• Modularize the main calculation steps in the pipeline.
  • Coordinate value sampling
  • Path length and moment arm calculations
  • Sample filtering
  • Polynomial coefficient fitting

Regarding modularization, each step could have their own internal/private class to improve maintainability and make unit testing easier. These classes could possibly be moved into osimCommon proper to facilitate this.

New features and performance improvements

• Detect moments arms for a path based on model topology, rather than (currently buggy) moment arm calculations. (Add AbstractGeometryPath utility for detecting dependent coordinates #4347, PolynomialPathFitter: use topology-based moment arm detection #4348 ).
• Reduce parallelization bottlenecking.
• Support generating a grid a samples without needing a coordinates trajectory.
• Make sample filtering optional, and improve current filtering approach (removes too many samples in some cases).
• Support fitting point-direction-based representations of paths to allow applying body forces via function-based paths.
• Reduce default fitting tolerances, which are probably a bit too aggressive.