Constrained internal optimization method should intelligently switch between cartesian and internal

Currently, the biggest hurdle to efficient transition state searches within Turbomole is Turbomole's tendency to derive linearly dependent internal coordinates.  Turbomole then crashes when it tried to optimize in cartesian coordinates with a fixed internal.

Implementing a better switching algorithm is probably the highest priority goal currently.
