- No need for nqp::eqaddr directly in the old tests.
- We need to test for specializations and classes' role_typecheck_list
  in addition to parametric roles. Surprise! We have more work to do
  here.

- Updating a role typecheck list should delegate to another role
  typecheck list at one point or another.
- The role typecheck list should become part of the type cache for any
  given MRO-ish type.
- The role typecheck list should carry a containerless `VMArray` of
  roles, as is the case with the MRO which, combined with the type
  object itself, forms the type cache of a supportive kind of type.
- Because a concrete role's instance is optional as `RoleToClassApplier`
  depends on it, such a nullish value should not become part of the
  roles list.
- Due to the simplicity of the various NQP HOWs' typechecks, each can
  carry a definitive type cache.
- Curried roles need typechecking in order for Rakudo to build with each
  of these changes.

This is more or less a direct translation of Rakudo's `roles`
metamethods as applied. Besides new parameters, the key difference is
`NQPClassHOW.roles`' `:$local` parameter, which (thankfully) was
required before. It should be capable of matching Rakudo's `0` default
because of that.

This is a recursive algorithm, echoing problems with coercions back in
the day. There's a better approach to this.

This reverts commit 533acf9.

`Metamodel::C3MRO` now expects a `:concretizations` parameter, but we
lack the `Metamodel::ConcreteRoleHOW` it wants from whatever carries it;
`:roles` is based off this as well.

A backport of Rakudo's. We lack its efface method here since we lack
`^parents(:excl)`.

Backport of Rakudo's changes. Err against recursing by depending on the
MROs of parents instead.

Include the specialization, not a duplicate of the role that was
specialized.

For consistency with every other HOW.

Backport of Rakudo's changes. `NQPConcreteRoleHOW` follows MRO ordering
for consistency with `Metamodel::ConcreteRoleHOW`.

Backport of Rakudo's changes. Nothing's hidden in nqp, so nothing to
`:excl`.