Semitokens versus tokens in sorted containers

[StephenVavasis]

A "token" for a sorted container is a tuple (container,s), where s is called a "semitoken" and is just a wrapped integer. Many of the functions in the sorted container suite take or return semitokens rather than tokens because in pre-1.5 Julia, tuples with a mutable element needed to be passed around on the heap instead of the stack. According to this thread in discourse: https://discourse.julialang.org/t/arrays-of-structs-tuples-with-mutable-members/45754/5 this limitation no longer applies. Semitokens still have a slight advantage over tokens in that they take up half the storage space, but otherwise, there does not seem to be a performance advantage of working with semitokens.

Question: should the package get rid of semitokens entirely? They are a bit kludgy but possibly people are using them in codes right now. (I am.) Or should the documentation be clarified but the kludge remains?

[oxinabox]

My inclination is towards removing semitokens.
Removing things is a breaking change so we need to do it before 1.0
adding them back is non-breaking and we can do that post-1.0 if it turns out that halving memory used was super important to someone.

[StephenVavasis]

Closed via #787