Snapshot tests for the assembly output

Context: #simd > Is there an Array copy cost? @ 💬

Since fearless_simd is trying to be extremely performant, this issue is a proposal for assembly snapshot tests.

A common approach to debugging performance with fearless_simd is to compile out some function and check that it's been appropriately optimized. My proposal is to lightly automate this with "goldens". We create a corpus of no_mangle SIMD functions that we compile out to various optimized backends. The output is then stored in a golden and compared.

Thus, we can manually validate how the Rust compiler and LLVM optimizes fearless_simd code. Any changes in compilation or code that modify the assembly would be caught.