tskit-derive may need further attention #920

#919 removed the use of proc-macro-error2 due to that crate being unmaintained.
To do so, we made use of syn::Error.
But in at least one case we neglected to work out the correct span for the source of the error, which should be fixed.

More generally, it would be nice to dig it a bit deeper and see how much we actually need to rely on external crates for the procedural macros.