block: optimize bvec iteration in bvec_iter_advance

There is no need to only iterate in chunks of PAGE_SIZE or less in
bvec_iter_advance, given that the callers pass in the chunk length that
they are operating on - either that already is less than PAGE_SIZE
because they do classic page-based iteration, or it is larger because
the caller operates on multi-page bvecs.

This should help shaving off a few cycles of the I/O hot path.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>