Implement `EllipseBounds::closestPoint` after #3990

https://github.com/acts-project/acts/pull/3990 changed the boundary tolerance to consistently use `closestPoint` for the computation. `EllipseBounds` are a bit special here and have been left out for now because there are a few open questions:
- Are the ellipse axes really defined as local vectors? To me it seems much easier to just provide the radii. This would also be less error prone.
- There is also an average phi angle in the bounds parameters - seems like this duplicates the vectors mentioned above?
- Implementing `closestPoint` seems straight forward for euclidean distance as soon as the points above are clarified.
- Implementing `closestPoint` using the mahalanobis distance is more complicated. But I think the ellipse can be transformed after decomposing the metric into another ellipse?