RDMA/qedr: Use rdma_umem_for_each_dma_block() instead of open-coding

This loop is splitting the DMA SGL into pg_shift sized pages, use the core
code for this directly.

Link: https://lore.kernel.org/r/9-v2-270386b7e60b+28f4-umem_1_jgg@nvidia.com


Acked-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>