This fixes some verifier errors when live out implicit defs are
coalesced with identity copies. Fixes some reduced testcases from
issue #38788 but doesn't solve the original failure.
I was surprised this seems to obviate the special casing in
analyzeValue that's been there since the subregister liveness support
went in.
If I understand correctly (well I'm trying to understand things), this is needed since we now want to treat the IMPLICIT_DEF as an ordinary value. And some previously computed ValidLanes values would be incorrect as it has treated the IMPLICIT_DEF as producing undefined values so we originally did not care about those lanes?