feat!: change TraceWith trait bounds to be on the field types and not on generics when deriving Trace

[bluurryy]

NOTE: This is a breaking change!

Currently deriving Trace on a struct like...

struct GenericStruct<T> {
    fn_ptr: fn(T) -> T,
    phantom: PhantomData<T>,
}

...expands to...

unsafe impl<
    __V: ::dumpster::Visitor,
    T: ::dumpster::TraceWith<__V>,
> ::dumpster::TraceWith<__V> for GenericStruct<T> {
    #[inline]
    fn accept(&self, visitor: &mut __V) -> ::core::result::Result<(), ()> {
        ::dumpster::TraceWith::accept(&self.fn_ptr, visitor)?;
        ::dumpster::TraceWith::accept(&self.phantom, visitor)?;
        ::core::result::Result::Ok(())
    }
}

Here there is a trait bound on the generic parameter T even though both the fields implement
TraceWith regardless of T. We should only care that the field types implement TraceWith, not T, so that
GenericStruct implements Trace even if T doesn't.

To implement this I've added a new dependency synstructure that delegates to the fields and collects trait bounds. I think it's worth adding this dependency because it massively simplifies the code and would allow easily adding field attributes in the future (like a trace(ignore) (I'd send a PR if this is merged)).

With this PR the derive macro expands to...

const _: () = {
    unsafe impl<__V: ::dumpster::Visitor, T> ::dumpster::TraceWith<__V>
    for GenericStruct<T>
    where
        fn(T) -> T: ::dumpster::TraceWith<__V>,
        PhantomData<T>: ::dumpster::TraceWith<__V>,
    {
        #[inline]
        fn accept(&self, visitor: &mut __V) -> ::core::result::Result<(), ()> {
            match *self {
                GenericStruct {
                    fn_ptr: ref __binding_0,
                    phantom: ref __binding_1,
                } => {
                    {
                        ::dumpster::TraceWith::accept(__binding_0, visitor)?;
                    }
                    {
                        ::dumpster::TraceWith::accept(__binding_1, visitor)?;
                    }
                }
            }
            ::core::result::Result::Ok(())
        }
    }
};

[claytonwramsey]

Thanks! I think this is a good idea. I'll shelve this PR for a little while as I wait to bundle more breaking changes (e.g. #81) but will merge soon.