Nimblephysics

- Fixes a segfault we were getting on `createShapeNode()` in Python
- Fixes an issue where we were detecting spherical joint centers in `MarkerFitter` that were worse than axis joint centers.

This is a few misc gradient bug fixes.

When fitting skeletons to markers in MarkerFitter, we've added support for adding a force field loss to encourage joints to stay at least a minimum distance from one another.

Also, several major stability improvements.

Summary of changes
- Fixed a bug in static trial optimization where decision variable vector length was not correct length, leading to memory corruption.
- Fixed bugs in `MarkerFixer` that were creating discontinuous marker traces and prematurely dropping marker frames.
- Added an option `DynamicsFitter::estimateFootGroundContacts()` to ignore checks that detect when a foot is over a force plate, which can be useful for GRF data that has missing or incomplete force plate geometry.

Summary of changes
- You can now specify a static pose trial in `MarkerFitter` using `setStaticTrial()` to add a cost term (with weight from `setStaticTrialWeight()`) to provide a prior to the marker fitting optimization based on the static pose joint angles.
- Added an option to disable `optimizeMarkerOffsets()` which is called during `DynamicsFitter::timeSyncAndInitializePipeline()`.
- For locked coordinates, `OpenSimParser` only set upper and lower bounds for the coordinate position. We now set equal upper and lower bounds for coordinate velocities and accelerations, which will prevent small "wiggles" in locked coordinates during dynamics fitting.
- Added support for ground reaction file headers from the Rajagopal model trial dataset.