This is my best attempt at making a decent chacha20 algorithm.

You need a machine with both `avx512vl` and `avx512f` flags. Other avx flags don't count.

To test this, run `lscpu | grep avx`

The full avx file is about 167.2% faster than no avx which is very signifigant.

The asm file, though, is only about 1.4% faster than the full avx.
- note the init function is about 50% faster but that isnt a majority of the work

So while I am happy to say that I can hand roll ASM better than gcc, it probably isn't worth it in most situations.

The actual benchmark I took on a 2 core Xeon Platinum 8175M AWS instance is in BENCHMARK.txt for anyone who is curious.