Incorrect branch weights being added around vector epilogue loops

[david-arm]

See PR #144682, which adds a test showing that we set the wrong branch weights in some cases, in particular in the vector epilogue iteration count check block where the branch weight suggests we always skip the epilogue.