Increase Instruction::raw_insn_T.small_insn size

Increase Instruction::raw_insn_T.small_insn size

By making this union .small_insn a uintptr_t, so it's the same size as
.large_insn, we can avoid some memory allocation without taking any more
memory in Instruction itself.  (PPC instructions are always 32-bit, so
there it's left as unsigned int.)

With this optimization on x86_64, the number of large_insn allocations
required is greatly reduced.  Many instructions are more than 4 bytes,
but it's less common to have more than 8 bytes.