I was reading over the gnome-class macro crate recently and I was having
a difficult time getting up to speed on the error handling of the crate
as it was using a number of patterns I hadn't seen before. I was also
wondering how only one `Span` was necessary to create an error when
multi-token spans were needed sometimes, which led me in the end to
start a refactor here.

The first point of refactoring was to ensure that multi-token errors
were handled. Although the `syn` crate has a `Spanned` trait it doesn't
actually work on stable Rust, so the errors coming out of gnome-class
today weren't compatible with being spanned over multiple tokens. I
added a convenience API in https://github.com/dtolnay/syn/pull/538 and
have started using that here to construct syn `Error` instances.

Next I opted to lift similar error handling from the `failure` crate
here as well. This was primarily the `bail!` macro as well as the
`format_err!` macro, used to quickly return and create errors with
formatted messages.

Using this macro I was able to replace most of the support traits in
`src/errors.rs` with local error handling and/or various other idioms.
Afterwards the crate (to me at least) feels much more idiomatic in terms
of error handling where `?` is still maximally used, returning an error
is easy to do, and there aren't too many traits/methods to discover when
learning how to work with errors.