This may fix the bug that's blocking Issue #2441.

Sadly, this exposes another ICE when trying to use the new version with Graph500

conservatively do it always for now

also: remove "auto-mode-matching" for implemented interfaces, as it
is complex and interacts poorly with classes

cc #2263

this is not clearly going to work due to monomorphization, which convverts
each enum<T> to a distinct "type" from POV of the shape code

fn subtyping is starting to work, but it's exposing bugs in the
code which checks for iface conformance, which doesn't properly
skolemize the self region

It's not that related to metadata, but metadata needs it and it will
probably be useful for doing dynamic loading.

The poststate should be one where all predicates are assumed false,
rather than the unchanged prestate.

Closes #2374

Closes #2311

note: you can still move from copy/move mode args

this is hacky, but shape code is going away anyway and I didn't
want to invest too much effort into it

Reachability was considering nested items to be unreachable, which
was causing the bug in #2383. Once I fixed that, I also had to make
impl::make_impl_vtable instantiate methods where necessary, before
calling monomorphic_fn.

Closes #2383

Also added some asserts and logging to trans.

Modified graph500 to use the shared_arc, but this unfortunately doesn't work due to #2444.