Elide lifetimes in derived functions

[drivasperez]

Closes #211.

The bug was caused by user structs and enums using the same lifetime ('a) that was used by some of the derived functions (help, url, code) in their signatures. So for the following example:

use miette::Diagnostic;
use thiserror::Error;

#[derive(Debug, Diagnostic, Error)]
#[error("foo")]
#[diagnostic(help("foo"))]
pub enum Foo<'a> {
    Bar { bar: &'a str },
}

you would get derived code that looks like this:

use miette::Diagnostic;
use thiserror::Error;
#[error("foo")]
#[diagnostic(help("foo"))]
pub enum Foo<'a> {
    Bar { bar: &'a str },
}

impl<'a> miette::Diagnostic for Foo<'a> {
    fn help<'a>(
        &'a self,
    ) -> std::option::Option<std::boxed::Box<dyn std::fmt::Display + 'a>> {
        #[allow(unused_variables, deprecated)]
        match self {
            Self::Bar { bar } => {
                std::option::Option::Some(
                    std::boxed::Box::new({
                        let res = ::alloc::fmt::format(
                            ::core::fmt::Arguments::new_v1(&["foo"], &[]),
                        );
                        res
                    }),
                )
            }
            _ => std::option::Option::None,
        }
    }
}

Which is a compile error because 'a is defined twice. However, it seems like you can just elide the explicit lifetime in the generated functions, which gives you code that looks like:

impl<'a> miette::Diagnostic for Foo<'a> {
    fn help(&self) -> std::option::Option<std::boxed::Box<dyn std::fmt::Display + '_>> {
        #[allow(unused_variables, deprecated)]
        match self {
            Self::Bar { bar } => {
                std::option::Option::Some(
                    std::boxed::Box::new({
                        let res = ::alloc::fmt::format(
                            ::core::fmt::Arguments::new_v1(&["foo"], &[]),
                        );
                        res
                    }),
                )
            }
            _ => std::option::Option::None,
        }
    }
}

So all appears to be well. I changed most of the derived tests to use a 'a lifetime while trying to work out where it was breaking. It seems like this is a case that should be covered but maybe in its own specialised unit test rather than scattered through the others?