According to the documentation, 32-bit x86 builds do not have the
64-bit _BitScan{Forward,Reverse}64 intrinsics.

Given these, we can remove the code to check for the zero case.

This patch includes an intrinsic implementation for these functions
when the compiler is GCC, Clang, or MSVC--or if the compiler claims
to be one of those.  Otherwise, a naive implementation is used.

Tests are included for all of these functions, which turned up a
possible problem: according to the gcc documentation, __builtin_ctz
and __builtin_clz give an undefined result when their inputs is
zero.  I was not able to persuade myself that we always called them
with a nonzero argument.

remove main() testing code from timeout.c and tweak Makefile so we can compare different wheel sizes with the benchmark framework

re-add one of our old bit fill operations, as it turns out it was actually needed (probably shouldn't have worked on that code in a sleep-deprived state)

add version routines and TIMEOUT_PUBLIC prefix to exported routines so applications can optionally compile completely statically (by defining TIMEOUT_PUBLIC to `static'

add invariant assertion code, and comment our source a little better (this really needs a formal description)

use timeouts_ prefix instead of timer_, which is in the POSIX namespace (they get all the good prefixes!

work backwards as well as forwards when calculating mask to overcome cummulative error when step is greater than period