Remove 'a lifetime on GraphQLError type

[jkoudys]

Including lifetimes on error types introduces huge overhead and greatly reduced flexibility for their use. Especially with the ? now in common use, intoing the Error type based on the Result returned by the function becomes extremely difficult when you have to preserve its lifetime.

I'm happy to write a PR myself, but want some feedback on the best place to remove it. Right now, it's simply the Token<'a>, which means juniper::parser::ParseError<'a>, which means juniper::GraphQLError<'a>, which bubbles up to every single caller. This is way too much insight into the inner workings of juniper than should exist on a public error -- I just need a nice message if there's a problem parsing, not the whole token. Right now it makes this impossible without separately mapping the error to remove it:

    juniper::execute_async(
        query,
        None,
        &RootNode::new(Query, Mutation),
        &vec![(
            "uuid".into(),
            InputValue::scalar("04310DD5269111EAAD050242AC120003"),
        )]
        .into_iter()
        .collect(),
        &ctx,
    )
    .await?

because now &RootNode::new(Query, Mutation) is borrowed but the await?'s result Error consumes it.

The obvious options:

• remove the Token completely and return some error info we build
• clone the Token (it already impls Clone) so we can pass back a Token<'static> instead
• open to suggestions..

[btielen]

While working on an integration with Axum, I also encountered this problem. For me, the lifetime specifier in juniper::http::GraphQLBatchResponse<'a> makes it really hard to work with the response.