An experimental idea to see if we can whole-sale cache the results of certain effect computations, when we can be confident that their repeated invocation should result in the same outcome. The idea is that a first interpretation of the effect stack (e.g. during a validation or dry run) can produce this memoization cache and repeated interpretations of the same effect during the same program session can just lookup results from th cache.

A good example is parsing of SQL files from disk. We'd want to do it once and have much faster resolution on repeat runs.