ENH: Update neb.py to improve the idpp implementation's speed and flexibility. (Fixed)

While investigating this method to NEB, we have found a couple changes that are useful to make. Using BFGS makes this code extremely slow for large systems and is totally unneeded as MDMin handles this system quite well. We also added an option to limit the number of steps the IDPP method is allowed to use to improve the NEB. It appears all systems converge in < 25 steps normally, so 100 is a very forgiving default number. Even difficult systems such as Zeolite frameworks with a large number of degrees of freedom converge quickly with this change.

Previous MR was really messed up so I decided to close that one and open this one. I have fixed the tests @jensj . One test didn't actually have a good enough convergence to justify its precision and the other system was a hard system for MDMin, which is atypical as we have been seeing it is unneeded to use BFGS in almost all systems, except when it is obvious that MDMin has failed spectacularly.