Improve the ergonomics of parsing and peeking

[udoprog]

This is accomplished through a bunch of new codegen in tools/generate, a couple of token markers and improved parser buffering.

The two primary macros added are:

• T, which will produce the ast type for the given fragment. For example, T![==] gives ast::EqEq.
• K, which produces the token kind of the given fragment. For examples, K![==] gives ast::Kind::EqEq.

These are less error prone and more compact than their ast::* equivalents because they visualize what the produced ast means, like:

p.parse::<T![async]>()?;
p.parse::<T!['(']>()?;

Instead of:

p.parse::<ast::Async>()?;
p.parse::<ast::OpenBrace>()?;

The Parser has also been modified to for improved peeking using Peeker and its associated nth method, which allows for peeking as many tokens as necessary forward using a circular buffer.

New Peek impls look like this:

impl Peek for Expr {
    fn peek(p: &mut Peeker<'_>) -> bool {
        match p.nth(0) {
            K![async] => true,
            K![self] => true,
            K![select] => true,
            ast::Kind::Label(..) => matches!(p.nth(1), K![:]),
            K![#] => true,
            K![-] => true,
            K![!] => true,
            K![&] => true,
            // and so forth...
        }
    }
}