[multibody] Speed up inverse dynamics by exploiting new templatized code

[sherm1]

[WIP] Don't review yet.

What's here

• Removes passed-in context from some calls in favor of passing in positions and velocities. (This is moving in the direction of standardizing the parameter lists so they can be replaced with preloaded structs.)
• Added inlines calc_A_FM() and calc_tau() to all mobilizers and replaces calls to the equivalent virtuals.
• Reimplements the virtuals using the inlines.
• Updates both CalcSpatialAcceleration and CalcInverseDynamics because the latter depends on the former.

---

This change is