Better loglikelihood for GLMM

[palday]

Amongst other things, this makes it so that the show(::GeneralizedLinearMixedModel) method works for models with a dispersion parameter.

(If it's not obvious from the commit history, this is me stripping off part of #291 so that at least some of those improvements make it in sooner rather than later.)

[codecov]

Codecov Report

Merging #419 into master will increase coverage by 0.07%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #419      +/-   ##
==========================================
+ Coverage   94.31%   94.39%   +0.07%     
==========================================
  Files          23       23              
  Lines        1637     1642       +5     
==========================================
+ Hits         1544     1550       +6     
+ Misses         93       92       -1     

Impacted Files	Coverage Δ
src/MixedModels.jl	100.00% <ø> (ø)
src/generalizedlinearmixedmodel.jl	87.24% <100.00%> (+0.68%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 183c94a...db006fd. Read the comment docs.

[dmbates]

Sorry not to review this previously. I just noticed now that you asked for my review.

At some point we should revisit the decision to use NaN for the value of sdest when there is no dispersion parameter. It seems that in modern Julia we should consider missing so we can extract a value of type T with coalesce(sdest(m), one(T)).

Thanks for straightening out the use of ϕ in loglikelihood.

[palday]

So we had actually revisited that decision in a previous PR (#418, on GLMM parametric bootstrap) -- sdest and varest return missing for families without a dispersion parameter, but dispersion returns 1 for compatibility with GLM.jl. Over- and underdispersion are things I would like to think about there as well, but that comes after the dispersion parameter fits even work consistently.....

I've now merged those changes back into this branch and am re-running the tests.