Make bit ops use u32 for shifts

[fjarri]

A part of #268:

• make all shift arguments in various bit shifts use u32
• the corresponding constants (e.g. BITS) and methods (e.g. bit_precision()) return u32. Note that the methods and constants that are used for indexing (LIMBS and BYTES) are still usize.
• added some methods to CtChoice (working on u32 arguments) and normalized their names.

Notes:

• I also added inline to all the CtChoice methods which bumped performance quite a bit in my tests (tens of percents in various benchmarks).
• I tried using u8 and u32 in CtChoice, but it seems to marginally decrease performance, so Word stays for now.

[tarcieri]

I'm definitely in favor of this since I found our use of usize oddly inconsistent with core/std. Also as-is subtle has no trait impls for usize which meant casting back to u32 from usize (which is not guaranteed to be lossless).

Looks like it needs a rebase though.

[fjarri]

Rebased