Add optional simd feature to optimize adler32 and miniz_oxide

[cjgriscom]

In a benchmark that generates many small PNGs, half of the runtime is split between miniz deflate, and the other half in png_pong::zlib::adler32. Replacing the built-in adler32 function with the optimized simd-adler32 crate makes the runtime of the adler32 function almost negligible. This PR adds the 'simd' feature to select between the built-in function and the external crate. It also enables simd optimizations for the miniz_oxide crate.

[AldaronLau]

@cjgriscom Nice, thanks for the PR!

I'm thinking since it looks like that crate has a fallback non-simd implementation for unsupported architectures, it would be preferable to depend on it unconditionally, rather than gated behind a feature. Then png_pong could get rid of the slow implementation it has.

[cjgriscom]

@AldaronLau I can't think of a reason that wouldn't work. I believe miniz_oxide leaves it behind a flag because of their no-unsafe policy, but if you don't care about that I will rewrite the PR to enable it unconditionally.

[AldaronLau]

@cjgriscom Since there are already a few instances of unsafe in the pix dependency for optimizations, I wouldn't be opposed to having unsafe in another dependency for a specialized optimization.