Clean up Fit tests that use random noise

[martyngigg]

Some fit function tests are actually more stress tests and they add random noise to test whether the fit works. The problem is that the random number sequence is not the same and this causes them to fail every so often when the DELTA in the test is not large enough. We need to be able to guarantee an answer for the unit tests so that we can be confident about what they are doing.

A simple change to a fixed random sequence will be fine

See http://trac.mantidproject.org/mantid/ticket/5734 which is to have better robustness testing of the fitting engine.

[martyngigg]

This issue was originally trac ticket 5736