[hunspell] learn replacements provided interactively by the user

[drahnr]

Summary

Currently the same custom replacement has to be typed over and over again, for no reason.

It would be better to learn one instance and then re-run the checkers with the newly recorded item for the current doc comment and all remaining ones.

Will require some re-thinking,one of the following two:

check all items...

• lazily in interactive mode, i.e. by moving to an iterator only design.
• and recheck on every learned item, requires to record all decisions of the user made before

Relvenat files: interactive.rs ,hunspell.rs

[drahnr]

One optione would to use dashmap to store the user selections and skip follow up corrections. This would be significantly easier if most of it would be done async on demand I think.

[blyxyas]

This would be so appreciated! In big projects there are lots (dozens and dozens) of instances where words like "config", "JSON", or "IDE" are flagged. It would be nice to keep track of which words have already been discarded as typos and just skip them next time.

[drahnr]

Work has started, but given a surge of lack of time, it's going to take some more time #296 - note that it's a bit more complex than anticipated and requires some additional refactor to work consistently.