improved hypothesis coding for contrasts (#172)

* widen types for contrats/hypothesis coding and use labels

* helper functions for computing hypothesis matrix

* separate type params for contr and hyp mats

* dict(label=>hypo_vector) constructor for hypothesis contrasts

* more hypothesis coding tests

* use symdiff to check difference

* one more test

* ContrastsMatrix: type for termnames != levels; constructor cleanup

* copy dataframe for missings test in contrasts

* reverse

* pretty hypothesis matrix with rationalize

* update docstring for AbstractContrasts

* use baselevel instead of .base; require labels for hypotheses

* don't need union, use AbstractVector for fields of hypo coding

* Apply suggestions from code review

Co-Authored-By: Milan Bouchet-Valat <[email protected]>

* use in(... fieldnames(T)) instead of hasfield

* warn, don't error for now.

* code review: error messages, formatting, docstrings

* remove test for warning since it's not clear when it triggers...

* T -> U

* revise docstring

* hypothesis matrix in AbstractContrasts docstring

* document kw constructors and note defaults

* docstrings, refactor prettify, use same orientation

* test hypothesis_matrix, sum inside cols instead of rows for intercept

* use sum instead eachcol

* more code review

Co-Authored-By: Milan Bouchet-Valat <[email protected]>

* define specific baselevel methods for types with .base field

* use hypothesis_matrix in docstring

* todo: termnames

* use tolerance= instead of pretty= in hypothesis_matrix

* final code review

Co-Authored-By: Milan Bouchet-Valat <[email protected]>

* fix up constructor

* dict -> Vector{Pair}

* Apply suggestions from code review

Co-Authored-By: Milan Bouchet-Valat <[email protected]>

* bump patch version

* add @nalimilan's suggested types for clarity

Co-authored-by: Milan Bouchet-Valat <[email protected]>

Author: kleinschmidt