replace the use of GaussJordan matrix inversion

[eugene]

The fitting functions need the inverse of the 'A' matrix to find the correct
errors. The LUD version of the code does not include matrix inversion. We MUST
have the errors calculated, so I have reset this code to use the psGaussJordan
function rather than the LUD version. George reverted to the LUD code since
some tests were failing. If this continues, we have a few options:

1) understand why the tests are failing and fix the problem (eg, is the input
matrix not actually invertable?)

2) code the LUD version of the matrix inversion

3) catch and warn on GaussJordan failures -- this is valid if the error is the
result of the input data.

In any case, the tests which are run to validate this code should really include
validation of the error calculation.

[Paul Price]

I believe the polynomial fitting functions now use exclusively G-J to solve the
matrix system.