Investigate representation issues for enum-like types

[jasoncarr0]

It was noticed with type spliting https://github.com/MLton/mlton/pull/285 that there was a performance penalty for boolean-like types that were changed to another type. A very naiive translation to Word32's was able to improve their performance when split-types-bool always was enabled, indicating that there may be some cost to the current representation (such as their size, or perhaps bit order).