rework fusion framework based on immutable named types

[mccanne]

This commit reworks downcast, upcast, and fuser to presume
that named types are immutable. The changes for immutability
are forthcoming but the fusion code now assumes this invariant.

This new design involves fused types carrying the named types
instead of stripping them and deferring them to the fusion
subtypes. This is key to allow the fusion runtime to manipulate
named types. When support for recursive types is added,
the recursive type will remain in the fusion and the concrete types
will be fused below. With this approach, a fused recursive type will
rarely need to be unfused.

This commit also tightens up the algorithms for upcast/downcast
so they should now be generally more reliable and can serve as
a model for implementing their vam counterparts.

[mccanne]

Now that we have named types fusing correctly and retained in the fusion type, I'm hitting a little problem where named type redefs cause a problem. Since we are getting rid of this with the forthcoming recursive types, I will work on a separate PR to disallow redefs of named types then rebase this work to that and continue on. It's getting close.

[mccanne]

The immutable-types PR creates a new set of problems with the used of serialized super data in the database metadata. This will be solved with abstract types so deferring that and instead, here, we are going to just change fuser to error on mutated named types and avoid that case in the tests.

[nwt]

Suggested change

	// We will soon maked such types immutable, but for now we just
	// We will soon make such types immutable, but for now we just

[nwt]

Suggested change

	if t, ok := t.(*super.TypeNamed); ok && named.Name == t.Name {
	return true
	}
	return false
	t, ok := t.(*super.TypeNamed)
	return ok && named.Name == t.Name