Avoid needless gather in fast_integer_divide lowering

[abadams]

fast_integer_divide did two lookups, one for a multiplier, and one for a
shift. It turns out you can just use count leading zeros to compute a
workable shift instead of having to do a lookup. This PR speeds up use
of fast_integer_divide in cases where the denominator varies across
vector lanes by ~70% or so by avoiding one of the two expensive gathers.

[steven-johnson]

Unrelated: since C++11 onwards default to promising thread-safe initialization of statics, we could rewrite this to be a bit terser like so:

Buffer<uint8_t> integer_divide_table_u8() {
    static Buffer<uint8_t> im(256) = []() {
        Buffer<uint8_t> im(256);
        for (uint32_t i = 0; i < 256; i++) {
            im(i) = table_runtime_u8[i][2];
            if (i > 1) {
                internal_assert(table_runtime_u8[i][3] == shift_for_denominator(i));
            }
        }
        return im;
    }();

    return im;
}

[abadams]

Good point. Will fix.

[abadams]

No, that's compile-time code.