Add a new `ParserExtra::State: Inspector` trait

[jyn514]

This allows using imperative concrete syntax tree parsers like rowan and cstree. In particular those libraries want to know every token that is parsed, and need to know when chumsky backtracks and reparses the same tokens again.

This adds the following new API surface:

pub trait Inspector<'a, I: Input<'a>>: Default {
    type SaveMarker: Copy + Clone;
    fn on_token(&mut self, token: &I::Token);
    fn on_save<'parse>(&self, offset: I::Offset) -> Self::SaveMarker;
    fn on_rewind<'parse>(&mut self, marker: Marker<'a, 'parse, I, Self::SaveMarker>);
}
pub struct SimpleState<T>(pub T);
impl<'a, T, I: Input<'a>> Inspector<'a, I> for SimpleState<T>;
impl<T> DerefMut<Target = T> for SimpleState<T>;
impl<T> From<T> for SimpleState<T>;

and additionally now requires ParserExtra::State: Inspector.

i didn't write any of my own tests yet but my own parser that hooks chumsky up to cstree does work: https://github.com/spreadsheet-lang/spreadsheet/commit/1b9731311aa97c1130c76de67966052909c35f75#diff-14962dfdd56397d99e07041d6337cb5b0327f4d7433d80fd3383955415014910R111

Fixes https://github.com/zesterer/chumsky/issues/96

[zesterer]

Looks like there are a few minor CI failures

[zesterer]

Thanks so much for your work on this, it's very appreciated!