Add a new text selection algo that should be a bit more reliable than the current one (just commented out), especially if the backend does not provide text areas in the logical order. Drawback: slightly slower.

Adapt both the algo's to work on untransformed text areas, but to return transformed areas.