Optional SIMD memrchr

[ncruces]

Continuing #580. memrchr is mostly used to implement strrchr, so that's what I test.

It has the advantage that we know the length and can access the entire buffer, so no undefined behavior for this one. I does do unaligned reads, however, so wasm_v128_load is used to dereference the pointer. It uses SIMD while there are 16 or more bytes to read, then fallback to scalar.

The only other notable feature is using clz rather than ctz.

[abrown]

Thanks! I've been fuzzing this, comparing output of the scalar and SIMD versions, and things seem to check out. I'll also note that just enabling __wasm_simd128__ (without this logic) seemed to do some amount of vectorization — something to compare against later.