With the new incoming types of gestures, it's important that we prepare a reasonable way to model them alongside the predictive-text engine. While we've long had a good setup in place for standard keyboard layers... none of the other uses cases have particularly great models.

Our existing longpress solution:

Yeah, that's not ideal. Surely we can do better.

So, to provide a good path forward for our existing longpresses and for newly-incoming multitaps and flicks, this PR creates abstractions for the minimal requirements of the key-distance algorithm at the core of the fat-finger algorithm. It also extracts the key-correction methods from ActiveLayer, allowing them to operate independently of any specific type - they'll now utilize the new abstractions.

By doing so, it becomes possible to build appropriate values for the new abstraction for our subkey menus, allowing true fat-finger calculations for subkeys. (A major improvement over simply maintaining the existing hack.) It also provides a baseline for further enhancements we may wish to consider for corrections involving multitaps and flicks. Note that none of the ideas in this paragraph are implemented within this PR, though - I've kept the scope limited to just the refactor needed to facilitate implementations thereof.

An extra benefit from this refactor - the key-distance types and algorithms at the heart of the fat-finger algorithm have been moved to input-processor, the same package containing the fat-finger algorithm. As input-processor's primary role is facilitating predictive-text and acting as the interface between keystroke-engine and predictive-text, it's the perfect place for them. (Not that input-processor is the perfect name, given the package's role.)

@keymanapp-test-bot skip