Fix control transform components and rotationOffset

[brianpeiris]

Description:
#3327 changed the way we calculate tracked controller pose:

• As a side effect, the code no longer updated the individual position and quaternion properties of the controller object. This prevents users and libraries (like networked-aframe) from reading those properties.
• It also removed use of tracked-control's rotationOffset property, this lead to hand-controls with incorrectly rotated hand models when using the Oculus Rift and Touch, as evident in the tracked-controls example.
• Lastly, it regressed standing matrix application for 3DoF controls -- i.e. standing matrix should only be applied for 6DoF controls, since the controller position is controlled by the arm model, which bases it off the head position.

Changes proposed:

• Decompose the tracked object's matrix in to its individual components
• Reintroduce use of the rotationOffset property.
• Only apply the standing matrix if the controller is 6DoF
• Add tests to cover these changes.

[brianpeiris]

I haven't tested this PR in a Daydream/GearVR with a 3DoF controller yet. We disable matrixAutoUpdate in init, so I'm not sure how this code could have worked for those controllers without an explicit updateMatrix. Perhaps it was actually broken. I'll test with a 3DoF controller when I get a chance.

[brianpeiris]

Confirmed. Daydream controller does not update without this.

[dmarcos]

Thanks for testing on it.

[brianpeiris]

@dmarcos After testing in Daydream, I found that this change was necessary as well. Just wanted to ensure it makes sense to you as well.

[dmarcos]

It makes sense, we can just make position (vrDisplay && pose.position)

[dmarcos]

Thanks for the tests

[dmarcos]

Thank you @brianpeiris 👍