Use 64-bit chunks on specific 32-bit CPUs

[nvzqz]

ByteSet is composed of [u64; 4] or [u32; 8] depending on the target architecture (determined in build.rs).

Currently, chunk size is based only on native register size. ByteSet should expanded be to use 64-bit chunks on 32-bit targets with meaningful 64-bit instructions.

Consider (and benchmark!):

• arm with neon target feature

[nvzqz]

For the sake of simplicity and making #8 easier, chunk size will only be determined by the target's pointer width.