relax SVN archive

Added some error checking to the specific_analyses.relax_disp.disp_data.average_intensity() function.

This is for better feedback to the user in case they have not set up their data correctly.

Added the 'test.seq' file from bug report #20873 (http://gna.org/bugs/?20873).

This is from Troels E. Linnet (http://gna.org/users/tlinnet). The bug report and link to
http://thread.gmane.org/gmane.science.nmr.relax.user/1452 explains the contents.

The file will be used to construct a system test to catch the bug.

The reference spectra can now be set in the relax_disp.spin_lock_field user function.

By setting the field to None, the reference spectrum for a fixed relaxation time period experiment
type can now be specified. This mimics the behaviour of the relax_disp.cpmg_frq user function.

Added a reference to the synthetic on-resonance R1rho test suite data.

The first ncyc1 data point now has a relaxation time period of zero, hence it can be used as the
reference for a fixed time period experiment.

Expanded the synthetic on-resonance R1rho test suite data.

The data now consists of a full set of dispersion curves for the M61 model.

Added some synthetic on-resonance R1rho data to the test suite.

This is in the form of Sparky peak list files containing two spin systems.

Reverted r19888 as the auto-analysis with just this model catches a bug users might encounter.

The command used was:
svn merge -r19888:r19887 .

The bug caught was fixed in r19889, but this system test now checks that it doesn't return.

.....
Fix for the Relax_disp.test_exp_fit system test.

Now this checks the fitting of exponential curves without using the auto-analysis - the
auto-analysis is an overkill.
.....

Model selection in the dispersion auto-analysis is only performed if 2 or more models are present.

Excluding the 'R2eff' model, if only 0 or 1 models are optimised, then model selection is skipped
and a warning is given. This avoids tracebacks in the model_selection user function.

Fix for the Relax_disp.test_exp_fit system test.

Now this checks the fitting of exponential curves without using the auto-analysis - the
auto-analysis is an overkill.

The error when selecting a non-existent model using relax_disp.select_model is now more informative.