relax SVN archive

Loosened all of the model-free tests so that the minimisation statistics are no longer checked.

The same result was being obtained on different platforms, different Python versions, and different
numpy versions, but the minimisation statistics were slightly different. The iter, f_count, g_count
h_count, and warning variables are no longer checked. This will stop users from thinking they have
a problem with their version of relax, after running the test suite.

The auto_analyses package modules are now imported by __init__ to force their existence.

The reference counting appears to not be needed in the C modules.

This allows the modules to be compiled under Python 2.4 or lower.

Scripts can now be run again under Python versions 2.4 or lower.

Fix for the initialisation of the Bruker_import class.

This allows this module to be loaded in Python 2.4.

The runpy module is only imported if it exists.

Added a check for the runpy module.

Fix for the back_calc_I C function so that it works under Python 2.5.

Added wrapper methods to the relaxation curve-fitting specific code for the new C modules.

These allow the parameter numpy array from minfx to be converted into a Python list prior to sending
it into the C module.

Removed numpy from the relaxation curve-fitting C module code.

This now expects Python lists as arguments. This should help with incompatibilities between
different Python versions, as this seems to be due to the numpy C interface. See for example
the thread started by Min-Kyu <min-kyu dott cho att vanderbilt dot edu>
https://mail.gna.org/public/relax-users/2011-02/msg00000.html (Message-id:
<001f01cbd24f$350042d0$9f00c870$@cho@...>).