Check all model & guess fitting functions

[jackbdoughty]

Check that in all model & guess functions, that they behave for all inputs e.g must check all for flat data as this is most likely to cause problems (div by 0 errors).

[Tom-Willemsen]

For the guess functions, if they really can't guess a sensible value for a parameter (e.g. guessing the centre of a gaussian with amplitude zero) then make them fall back to very simple guesses - e.g. always guess centre of gaussian is in middle of x range, guess a height of 1, etc... This is still "better than nothing" as a guess for the optimizer.

[Tom-Willemsen]

For a LinearFit we were seeing a consistent RankWarning on the first point measured.

Suspect this is because at some point, we tried to "guess" at a linear fit with just one point. Catch this case and don't try to guess for the very first point.