Add x86 intrinsics support for sha1 and sha2

[newpavlov]

Written using #90 and the software implementation as a reference.

This PR also simplifies handling of ARM backend and deprecates asm-aarch64 feature.

[dignifiedquire]

@newpavlov I agree, your code looks much nicer :)

Some thoughts for my specific use cases

(1) Bringing back dynamic detection would be great, often times the code will be built once and then run on different cpus in my case.
(2) now that input_blocks is a thing again it will be more efficient to process multiple blocks at once, like I am doing here: https://github.com/filecoin-project/rust-fil-proofs/blob/master/sha2raw/src/sha256_intrinsics.rs#L12 Looks like you already did this
(3) currently all implementations do have some overhead and additional padding in my measurements, which is why I ended up implementing this "raw" construction https://github.com/filecoin-project/rust-fil-proofs/blob/master/sha2raw/src/sha256.rs#L25 which avoids going through the block_buffer and can only process appropriately sized and pre padded chunks. It would be great to find a way such that passing in slices of matching data could have the same speed.

[Fleshgrinder]

I second the dynamic detection request. This is important for software that is installed on many different systems.

[newpavlov]

Bringing back dynamic detection would be great

The code is written with dynamic dispatch in mind, I just need to write CPUID detection using getrandom code as a reference (probably I will move it into a separate crate).

currently all implementations do have some overhead and additional padding in my measurements

If you don't need padding, then I think the easiest way will be to use the compression functions directly, which can be accessed after enabling the compress feature. I guess we also could add imports for initialization constants.

[newpavlov]

Runtime detection is back! If you interested, it's powered by the cpuid-bool crate.

@linkmauve
Can you verify that crates are not broken on ARM?

[dignifiedquire]

very nice, thank you!

[jethrogb]

This code breaks platforms that don't support CPUID (e.g. SGX). You should be using https://doc.rust-lang.org/std/macro.is_x86_feature_detected.html, which is the standard way all platforms support feature detection.

[newpavlov]

Note that is_x86_feature_detected is not available in libcore and the linked PR proposes changes to change it. See #183 for discussion as how to fix SGX.