Random notes: performance

[LukasKalbertodt]

I'm currently working on another performance-critical project that uses this crate. I'm inspecting generated assembly code and noted some possible improvements to this crate. Before I forget all of this, here is a random list of notes:

• Using bitvec[index] is bad: because that crate uses the static TRUE/FALSE trick to in the [] operator we have a branch where there shouldn't be one. Use get instead or something like that
• First indexing the bitvec and then the data vec does two bound checks. That's bad. Everywhere where we use has_element_at (and it returns true), we should use get_unchecked or so.
• Setting or resetting a bit (done via & or | with a mask) is not optimized well. This should (probably be done) with the BTS and BTR instructions on x86. Have too look into this.

(This issue is rather a reminder to myself...)