Infer trait bounds for generic derives

[epage]

Building off of #2769 / #3023, we should allow inferring trait bounds.

So instead of

#[derive(Parser)]
struct Opt<T: Args> {
    #[clap(flatten)]
    inner: T,
}

The user does

#[derive(Parser)]
struct Opt<T> {
    #[clap(flatten)]
    inner: T,
}

like they do with when deriving Copy, Clone, serde types, etc.

See https://serde.rs/attr-bound.html for some inspiration on issues to deal with.