Refactoring for lazy creation

[tgross35]

Some dictionaries are too big, e.g. Hungarian is >1GB RAM with our current setup.

To do the below, we will need to split our affix computations into create_word, our current format that makes the words in advance, and test_word that can see if an affix applies.

It would be best to control how lazy or eager we want this to be. For example:

• Always lazy: use test_word for everything, no internal state
• Dynamic: precompute nothing, cache results as we create words
• Fully eager: precompute everything and store it (like we do now)

[tgross35]

Probably want to overlap this with #45 somehow