Also special case int-aligned scalars on 64-bit machines.
On a 64-bit machine hashing an array of pointers is now ~5.95x faster
with LDC2 and ~8.54x faster with DMD, and hashing an array of ints is
~3.34x faster with LDC2 and ~8.12x faster with DMD.
This is appropriate because HashArray in tanya.container.entry uses
prime numbers instead of powers of 2 for its number of buckets so there
is no pitfall if the hashes are all multiples of some power of 2.