relax SVN archive

Big refactorisation of the value setting code. Expect breakages as the changes aren't complete!

The parameter setting set() base class method has been shifted from the specific code to
the generic_fns.value module (and has been converted to a function). This method simply sets spin
specific functions and hence has been renamed to the set_spin_params() function.

Converted the maths_fns.jw_mapping module from Numeric to numpy.

Removed the num_instances() method from the specific_fns code as it is now part of the base class.

This method was moved to the base class in r4804 and the change was explained in a post by Edward
d'Auvergne at:
https://mail.gna.org/public/relax-devel/2008-01/msg00063.html (Message-id:
<7f080ed10801170126p7d8f5109ib838a894f4f979a8@...>)

More fixes for the unit tests of the generic_fns.value.partition_params() function.

The current data pipe needed to be changed back to a model-free pipe.

Removed some debugging code which is now covered by the unit tests.

Fixed all of the unit tests of the generic_fns.value.partition_params() function.

Removed some useless code in the partition_params() function.

Because of the main if-else loop, this code will never be executed!

Updated the generic_fns.value module functions set() and partition_params() to be more general.

The updating is not complete, but works for most of the current codebase. The assumption of a data
analysis type having a diffusion or other tensor has been removed. This has been replaced by the
concept that the params are now partitioned into those that are spin specific and those that are
not. The diffusion tensor stuff will be moved into the model-free code.

Removed the grid search from the 5-state xz model system test.

This has been replaced by a call to value.set() as the grid search is too expensive.

Bug fix for the N-state model disassemble_param_vector() method.

The last probability for state N was being incorrectly calculated!