RFR: sh/jdk11 assert failure because IfNode input is a Phi

[Roland Westrelin]

> This seems to be partially backporting this change, right?
>   8213381: Hook to allow GC to inject Node::Ideal() calls

Yes.

> We would need to pick that up in 11.0.6 to minimize upstream differences. The current patch seems
> okay for sh/jdk11, as it is pretty straight-forward. If we want to be extra-clean, I'd wait for
> 8213381 backport to arrive via 11u.

I'm fine with that but we will have to remember that we need to undo the
previous fix too.

Roland.