Improve WMR pose conversions

This comes after discussing some pose corrections needed for hand tracking like sending the SLAM-tracked IMU pose to the middle of the eyes, or the left cam pose to the same place.

Hopefully, the way poses/transforms are handled now is clearer. The idea is to keep everything in WMR coordinates as much as we can, and only use a handful of transforms (precomputed in a specific function) for doing WMR-to-OpenXR conversions.