intel: use bitset in memory access lowering pass
The mem access lowering code uses an opencoded 32-bit bitmask. In the future to add vec16 support we need to expand this bitmask to 128-bit. So let's port the code to bitset now, so changing it later isn't as invasive.
This relies on !752 (merged) landing first, so marking it wip.
Edited by Dave Airlie