iommu/io-pgtable: Add fast page table mapper for ARMv8L

Certain use cases require performance that can't be achieved with the
general-purpose SMMU page table code.  By limiting ourselves to 4K page
mappings (no block mappings) and pre-populating the first and second
levels of the page tables up front, we can eliminate a lot of the work
needed for page table mapping and unmapping.

Add a performance-tuned io-pgtable implementation for ARMv8L page tables
that only supports 4K page mappings.  Any size can be mapped, but only
4K page mappings will be installed in the page tables.

CRs-Fixed: 997751
Change-Id: I5861270709675016988052360d196e0a16a0d103
Signed-off-by: Mitchel Humpherys <mitchelh@codeaurora.org>