separate vmm allocation units from kernel granularity

DR's vmm is still tying its own allocation granularity to that of the kernel when -vm_reserve is in place. When we have a large reservation we can ignore the 64K Windows granularity or the large page sizes of some AArch64 systems. The code will get more complex as once we're outside of -vm_size we'll have to deal with the kernel's units, but it may be worthwhile for the flexibility inside the reservation.