Improve `Block` with SIMD

Right now the backing type of `Block` is `[u8; 16]` as it was simple to start with. However, core arrays can not always take advantage of auto-vectorization/SIMD for important operations.

For example, on my machine, it currently takes ~3ns to compute `BitXor` between two blocks. If using `std::arch::x86_64::__mm128i` this would be <1ns. We can of course just convert the `[u8; 16]` to/from `__mm128i` for the operation, but the overhead of doing so erases any performance gains.

`Block` would ideally use an `u8x16` SIMD backing type depending on target architecture. Unfortunately `std::simd` is [not stable](https://github.com/rust-lang/rust/issues/86656) yet so we will have to provide implementations for the various architectures.

To briefly motivate doing this, after some quick profiling and napkin math it seems that garbling a circuit is currently bottlenecked on `BitXor`, not the CCR hashing!

For example, the AES circuit has 6500 AND and 30,163 XOR gates. Garbling this circuit currently takes ~350 microseconds on my machine. Garbling an AND gate requires up to 9 XOR ops. In total that is 88,663 XOR ops to garble AES, which adds up to 266 microseconds or roughly 75% of CPU time.