Macro diagnostics for missing overload or wrong arguments are confusing

[teor2345]

From a potential overloading user:

I also have a more general concern w/ the current approach wrt diagnostics, since "trait bound not satisfied for XYZ tuple" is likely to be confusing as compared with "no available overload for arguments abc, considered overload 1 but was not applicable because x, considered overload 2 but was not applicable because y, etc."

We should be able to generate better diagnostics by getting the macro to emit diagnostic::on_unimplemented and maybe diagnostic::do_not_recommend on some tuple trait impls.