RFR: Use BarrierSetC2::ideal_node() API for CmpPNode::Ideal()

[Roman Kennke]

Oops. Just now I've submitted the RFR. I did not know that we inject
::Identity() calls too. Yeah, that should go together. BTW, I think the
approach that I've chosen (see submitted RFR) is more generic in the
long run: it intercepts Ideal(..) not by requiring to override specific
::Ideal() calls, but where ::Ideal(..) is called in phaseX.cpp, which
allows GC to inject into *any* type of node, not only the ones that
we've prepared for it. Identity() can be treated the same way I think.
Want me to do it?

Roman

>> Should I take this upstreaming? I've already everything lined up:
>>
>> https://bugs.openjdk.java.net/browse/JDK-8213381
> 
> The LoadNode::Identity() change should also be upstreamed as a non
> shenandoah specific change. I intended to do that one and the
> CmpPNode::Ideal() together.
> 
> Roland.
>