Check for ->bd_invalidate moved from rescan_partitions() to the only caller
that
	a) needs that check and
	b) doesn't do it already.

Fixes the problem with BLKRRPART which doesn't want that check at all...