Inconsistencies of time to execute grouseticks test

[dmbates]

The optimum for the model in the grouseticks test is quite unstable. We had to comment out tests on things like the sum of the squares of the deviance residuals because we couldn't get a reproducible value.

I have been running tests on both the release version of Julia (1.12.4) and the beta release (1.13.0-beta1) and the amount of time that this test takes is unstable. It takes about 13 seconds with 1.12.4 and over 1 minute with 1.13.0-beta1 on my MacBook. We may want to look at this in more detail or we may want to remove the test.

[palday]

Do we have another model to test the Poisson functionality?

I've also noticed that there is a lot of instability in those fits.

Also, a 4x increase in fit time from a Julia version upgrade feels like something big has changed in LinearAlgebra. I'll try to bisect if I have time.