Derive a replacement variable

[nealrichardson]

Sometimes when you combine categories or do some other derivation, you don't want to see the input variable anymore. But it's useful to keep it around so that if more data is appended, the derived variable updates.

This PR creates a means of attaching metadata to VariableDefinitions when creating them that indicates that, when they are added to the dataset, that some other variable(s) should be hidden and this new variable should appear in its place in the order.

This should also resolve/standardize some of the awkwardness in #154

[codecov]

Codecov Report

Merging #162 into master will decrease coverage by 0.25%.
The diff coverage is 15.78%.

@@            Coverage Diff             @@
##           master     #162      +/-   ##
==========================================
- Coverage   89.21%   88.95%   -0.26%     
==========================================
  Files          92       92              
  Lines        5479     5497      +18     
==========================================
+ Hits         4888     4890       +2     
- Misses        591      607      +16

Impacted Files	Coverage Δ
R/add-variable.R	75.28% <6.25%> (-15.26%)	⬇️
R/combine-categories.R	97.66% <66.66%> (-0.56%)	⬇️

---

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 54fcd7c...0c2a098. Read the comment docs.

[jonkeane]

The code looks good to me. I'm a little bit hesitant that replace might sound misleading to the user as being a literal replace the variable (especially with respect to the alias). I can see that making the replacement variable have the same alias as the original could break things like streaming and appends. So this is probably as good as we can get, and really as good as people need. But maybe a note saying that replace=TRUE will and must have a new alias that is unique.