This is a good suggestion for incorporation into version 14 of Keyman
This is already supported; in version 14 we plan to extend this further.
Nuspell and Hunspell both add a significant dependency burden to Keyman which make them less than ideal for integration. So we are not planning to use them at this time, but may revisit this in the future.
The corrections algorithm currently uses a simplified distance algorithm to determine related words and uses key proximity for additional alternatives.
The primary documentation for the lexical model editor is still in development during the beta cycle; you can see what we already have at Lexical models