Fits are unstable

[mfisherlevine]

As shown by the variable results in issue 32, fits are quite unstable.

At most stages, despite being seeded randomly (at time of issue-filing == d568533) the intermediate results are identical to the displayed precision when displayed with --verbose (e.g. 300.23339285427346, which is lots), but the final result is not, so something is not right.

Fixing issue 32 will fix this for some cases, of course, but will simply cement others into a bad part of the phase space, depending on the seed and the input data, so this is a different issue entirely.

Warnings showing this include outputs such as Minimum chisq is on the edge of the exploration grid. as well as being evidenced by getting different results from the same input.