emmeans and all pairwise comparisons

[palday]

The math for estimated marginal means (aka least-square means) is essentially the same as for effects in the basic cases. (Life is a bit more complicated if we consider something like "emtrends", especially in the presence of interactions instead, of instantaneous measures.) This provides some full service functionality for computing these.

• [x] implement emmeans
• [x] implement empairs
• [x] add test statistics?
• [x] add docstrings
• [x] add docpage
• [x] note limitations of empairs when it comes to documentation etc.
• [x] note that the exact output format of empairs will likely change and be considered nonbreaking, i.e. empairs is experimental

[codecov[bot]]

Codecov Report

Merging #43 (4b44a2c) into main (445c8a8) will not change coverage. The diff coverage is 100.00%.

@@            Coverage Diff            @@
##              main       #43   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            2         3    +1     
  Lines           89       144   +55     
=========================================
+ Hits            89       144   +55     

Impacted Files	Coverage Δ
src/emmeans.jl	100.00% <100.00%> (ø)
src/regressionmodel.jl	100.00% <100.00%> (ø)

:mega: We’re building smart automated test selection to slash your CI/CD build times. Learn more

[palday]

I think so -- especially since the heavy lifting is still done by effects! which we have checked more closely. There's some subtlety with the implicit choice of weights, but that's something we can make more featureful in the future.