`RegularIrregular` also corrects regular forms but then its `message` is inaccurate

> that if statement that just replys "x" works on my machine killed me.

In this case "replys" is a wrong spelling of the plural and 3rd person singular present of "reply" and it does correctly suggest "replies", but the message says "Use the irregular form".

The linter should probably offer different `message`s depending on whether the word is actually irregular on the one hand vs having to follow certain orthographic rules on the other hand. I'm not sure what wording would be best though.

I asked an AI and it suggested
> Use the standard spelling rule: change 'y' to 'i' and add 'es' to form the third person singular.

Which is probably a decent jumping-off point.

Start by locating the `RegularIrregular` rule and how it constructs its `message`. Trace the handling of the `replys` example and inspect any nearby tests or fixtures. Done means regular spelling corrections receive wording that describes the orthographic rule, while genuinely irregular corrections retain the irregular-form message.

Written by the indexing model from the issue text.