Drop deviance and reformat lrtest table

[andreasnoack]

Summary

• Drop the deviance field from LRTestResult and the matching column from the printed table.
• Reorder/rename remaining columns to LogLik, p (number of parameters), χ², df (Δ degrees of freedom), p(>χ²).
• Center column headers, mirroring the style on the an/ftest branch of GLM.

Before

Likelihood-ratio test: 3 models fitted on 12 observations
────────────────────────────────────────────────────
     DOF  ΔDOF   LogLik  Deviance   Chisq  p(>Chisq)
────────────────────────────────────────────────────
[1]    1        -8.1503   16.3006
[2]    2     1  -7.9780   15.9559  0.3447     0.5571
[3]    4     2  -7.0286   14.0571  1.8988     0.3870
────────────────────────────────────────────────────

After

Likelihood-ratio test: 3 models fitted on 12 observations
───────────────────────────────────
     LogLik   p    χ²    df  p(>χ²)
───────────────────────────────────
[1]  -8.1503  1
[2]  -7.9780  2  0.3447   1  0.5571
[3]  -7.0286  4  1.8988   2  0.3870
───────────────────────────────────

Test plan

• Pkg.test() for the lrtest testset passes (73/73)
• Verify the jldoctest in the lrtest docstring matches the new format on CI

[andreasnoack]

I don't think it is common to talk about the residual or model degrees of freedom in a likelihood setting. You usually talk about the number of parameters in the model, here p. The difference in the number of parameters gives the degrees of freedom of the Chisq test statistic (provided testing). In addition, there is not simple relationship between the deviance and the LRT, so I'm removing the column to make the output more concise. Finally, a bit of styling changes. This is a bit related to JuliaStats/GLM.jl#628