Future PR16849 (D7204) introduces locking m_cs_chainstate to their version of InvalidateBlock. Our codepath for that is shared by ParkBlock, InvalidateBlock and FinalizeBlock, so having all three be members of CChainState (FinalizeBlock already is) makes sense.
Also made UnwindBlock private since it's not supposed to be called by just anyone, rather by stuff that knows what they're doing -, and moved FinalizeBlock implementation to be closer to the other two since that makes more sense.