Caching matrix after successful boundary crossing.

This MR caches the global transformation matrix in the navigation state at the moment of successful boundary crossing. The matrix will be then re-used as starting point for any attempt of re-location or computing distance to daughters. The net gain in FullCMS is of ~2% (out of the 4.5% hot-spot for matrix computation), while the memory cost is ~10 MB on top of a total footprint of 380 MB in full scalar mode.

The feature is used automatically when compiling VecGeom with the option USE_CACHED_TRANSFORMATIONS=ON