relax SVN archive

The grid search for the MMQ-type models now looks for negative chemical shift differences.

Fix for the legends in the Grace graphs produced by the relax_disp.plot_disp_curves user function.

Fixes for all of the Relax_disp.test_korzhnev_2005_*_data system tests.

These now start optimising at the solution found by cpmg_fit. All tests now pass.

Updated the Relax_disp.test_sprangers_data_to_mq_cr72 system test to pass.

The 'MQ CR72' model is now much more accurate due to a number of recent bug fixes.

Updated all of the cpmg_fit results to use the numeric 2-site CPMG model.

This also uses the modified cpmg_fit binary with higher accuracy.

The Relax_disp.test_korzhnev_2005_all_data system test no longer dumps files in the current directory.

Fixes for a number of dispersion system tests due to the higher accuracy of the tau_CPMG values.

This is required as the tau_CPMG values have been corrected to eliminate user input truncation
artifacts.

Eliminated the MODEL_LIST_CPMG_NUM variable.

This was far too specific and its misuse caused a bug in the target function of a number of
dispersion models.

Fixes for the relaxation dispersion system tests for the changed behaviour of the 'CR72' model.

The optimisation is slightly different as values are now always passed into the numpy.arccosh()
function.

Updated the cpmg_fit results for the CPMG data of Korzhnev et al., 2005.

This is using a modified binary wherein the gyromagnetic ratio and optimisation tolerances and
maximum number of iterations are far more accurate (to the same level as relax). The cpmg_fit
output has also been made more accurate by writing out the values to much higher precision.