Use inline assembly for md5

[johnmave126]

This is a md5 implementation using inline assembly instead of relying on cc and .S. This has the benefit of compiling and running on MSVC targets.

Correctness

I have run a custom version of md-5 pointing to this branch with cargo +nightly test --features asm and cargo +nightly test --features asm --target i686-pc-windows-msvc on a windows machine and passed the test.

Performance

On my Linux machine, which has a Intel 6700K
Before:

test bench_compress ... bench:         104 ns/iter (+/- 0) = 9 MB/s

After:

test bench_compress ... bench:         103 ns/iter (+/- 0) = 9 MB/s

No observable performance degradation.

Drawback

This change needs a MSRV bump to 1.59.0 where inline assembly was stabilized.

[tarcieri]

@johnmave126 with inline assembly I think you can just add it upstream to the md-5 crate

[johnmave126]

@johnmave126 with inline assembly I think you can just add it upstream to the md-5 crate

That makes sense. Forgot to check #45

[tarcieri]

Opened #55 to add info about that to the README