When initAnimGraph is called it asynchronously loads the Animation graph in the background.
If the model url is changed, or the Model is deleted in between the initial load and it's completion,
It's possible to access a bad Rig::_animSkeleton pointer.
The fix is to hold onto the _animSkeleton pointer via a weak ref.
* Domain settings version has been bumped from version 2.0 to 2.1
* Old domain settings for avatar scale will be auto-converted to avatar height
* Avatar code has been changed so that limitDomainScale() works with the new height limits
* Avatar getUnscaledEyeHeight() was added to C++.
* MyAvatar.getHeight() was added to JS.
Create a global registry to hold all the currently active Rig instances.
Use this registry and it's mutex to prevent accessing the rig after it has already been destroyed,
or is in the process of being destroyed on the main thread.
Make sure to explicitly set the _geometryToRigTransform in Rig::initJointStates().
This should fix small bounding volume when switching from the Gamebot avatar (which has 0.008 scaling)
to another avatar (which ususally have 0.01 scaling).
IK is no longer performed during capsule calculation, This fixes an issue with the mannequin avatar in the marketplace.
Specifically, it was causing the hips to rise during IK which would result in an incorrectly sized and offset capsule.
* AnimInverseKinematics: debug draw for secondary targets
* AnimInverseKienmatics: better clean up of ik target debug draw
* GeometryUtil: added findPlaneFromPoints()
* ViveControllerManager: external dependency on eigen
* ViveControllerManager: record history of left/right hand controllers
* ViveControllerManager: use history to determine user shoulder location for better calibration
* ViveControllerManager: pass defaultToReferenceMat by const ref to calibrate functions.
* CMake: added external depenency to eigen linear algebra library.
Also, A change in how the bone name to bone index lookup occurs exposed a bug
in Rig::computeAvatarBoundingCapsule(), basically it was not actually preforming IK,
and the ik targets were in the wrong coordinate frame. So when IK was actually
performed it would give bad results. This bug is now fixed.
