More efficient reverse_bits implementation.

[Parakleta]

In general the reverse_bits function should be generalised to work on whatever word type the BitVec object is using and then be applied to accumulator directly instead of each byte (assuming the bytes are reversed while assembling the accumulator value). Additionally bit switching should be cheaper than shifting. Consider:

fn reverse_bits<B: BitBlock>(block: B) -> B {
    let mut result = B::zero();
    result = ((block >> 1) & (!B::zero() / 3)) | ((block & (!B::zero() / 3)) << 1);
    result = ((block >> 2) & (!B::zero() / 5)) | ((block & (!B::zero() / 5)) << 2);
    result = ((block >> 4) & (!B::zero() / 17)) | ((block & (!B::zero() / 17)) << 4);
    result
}

I'm assuming !B::zero() / 3 etc. become compile time constants.