relax SVN archive

Fixes for the calc user function for the dispersion analysis.

This now does something logical for the non-R2eff models. The chi-squared value is now being
calculated and stored. Previously this was only calculating the R2eff/R1rho values for fixed
relaxation time period data for the 'R2eff' model and failing for all others. Now the pre-existing
_back_calc_r2eff() method is used to back-calculate and store the chi-squared value.

The dispersion data return_cpmg_frqs() and return_spin_lock_nu1() functions are now safer.

These specific_analyses.relax_disp.disp_data module functions can now be called when no data is
present.

The dispersion GUI analysis cluster_update() method is now thread safe.

This removes many error messages when running the dispersion analysis in the GUI, especially for Mac
OS X systems.

More loosening of the Relax_disp.test_hansen_cpmg_data_to_ns_cpmg_2site_star_full system test.

This is now for 64-bit Mac OS X to pass.

Caught a divide by zero in the specific_analyses.relax_disp.disp_data.return_offset_data() function.

This was identified by turning all numpy warnings to errors.

Loosened the checks for the Relax_disp.test_hansen_cpmg_data_to_ns_cpmg_2site_star_full system test.

This is to allow the test to pass on 32-bit Linux systems.

Fix for the lib.dispersion.cr72 module for early Python versions.

For Python 2.5 and earlier, the math.acosh() function does not exist. Therefore the numpy
equivalents are now being used.

Updated the Relax_disp.test_hansen_cpmg_data_to_ns_cpmg_2site_star_full system test.

This is for the changes of r21191.

Shifted all of the numerical dispersion code to use the internal matrix power function.

Instead of using the numpy.linalg.matrix_power() function, the relax
lib.linear_algebra.square_matrix_power() function is being used instead. This allows the code to
run on many older systems, as the numpy function is relatively new.

Removed an unused import.