Words of any bitwidth less than 64 in the semantics

[ordinarymath]

Restricted version of #499. It should just done in the way the translator currently encodes them, where if the bitwidth is less than 8 it is stored in a word8 and if the bitwidth is less than 64 it is stored in a word64.

[ordinarymath]

I do not intend for this to be a major change and it's not for performance and it can probably be compiled away in source_to_flat.