Benchmark 256-bit addition/subtraction

[alexcrichton]

Came up in the CG meeting today. Should benchmark what the performance gap is between native and wasm with 256-bit arithmetic using 128-bit addition.

[alexcrichton]

A small benchmark, also added to https://github.com/alexcrichton/wasm-benchmark-i128, shows:

benchmark	today slowdown	wide arithmetic slowdown
add256	82%	58%
mul256	2020%	101%

natively for me on Wasmtime (compared to native x86_64). So better, but not 100% of the way there.