Optimiser bugs

[shoubhikraj]

Compiling a list of bugs found so far in the optimisers, so that they could be fixed one-by-one.

1. CRFOptimiser - coordinate building does not seem to always work. Number of primitive might be larger than the expected number of degrees of freedom, and still not be enough after removing redundancies during the DIC generation procedure.
2. CRFOptimiser does not seem to work very well. The constraints are often not satisfied at convergence due to the energy convergence criteria being signalled earlier. This is probably because it is not choosing the constraint modes to maximise when there are other negative modes in the hessian.
3. Coordinate system has to be rebuilt in case one of the angles or dihedrals become close to 0 or 180 degrees (SVD did not converge: #143 (cont, supposedly fixed by #151) #261 )
4. HybridTRMOptimiser - removing last set of coordinates may cause infinite loop, this needs to be removed. The max_trust is also too high.
5. PRFOptimiser - accurate hessian is calculated in cartesian, but the cartesian to DIC hessian transform does not use the complete formula (which is fine for minimisation, but for TS search, accurate hessian is needed)
6. Optimiser plotting - coordinate has to be converted to cartesian before printing geometry.
7. Low-level hessian switches to high-level if no lmethod is available (?)
8. Improve trust update by separate algorithms for minimizer and TS search
9. Consistent convergence criteria - should include constraints
10. PRFOptimiser change to normal primitives from inverse distance primitives
11. Printing of convergence criteria should print signed energy change not unsigned