Add scalar/Neon/Neon hybrid for Keccak-x4

[hanno-becker]

This PR adds hybrid scalar/Neon/Neon implementations of Keccak-x4. Those implementations were first described in https://eprint.iacr.org/2022/1243.

We don't consume the implementations of that paper as-is, however, but auto-generate the code from clean de-interleaved assembly. As it stands, only the interleaved versions are added here.

[hanno-becker]

I haven't yet amended optimize.sh to reproduce the optimizations from de-interleaved code. As it stands, the optimizations are only implemented in example.py in the draft PR slothy-optimizer/slothy#65.

I'd prefer to do this in a follow-up, but could be convinced otherwise.

[mkannwischer]

Thanks @hanno-becker!!
It's getting rather hard to keep track of which implementation is used for each target. #141 should be addressed soon.

Code looks mostly good to me - some comments.
Here is the performance of keccak-f1600-x4 cycles

platform	before	after
A72	6196	5407
A76	3708	2607
A55	6012	6346
G2	3709	2608
G3	2078	1604

That looks great except for the A55 - there, you just want to use a scalar 1x implementation throughout. That's in line with the paper.
We'll need a special case for the A55 I guess. Probably better to do that as a follow-up PR.

Thanks for introducing the profile for the special case A55.

[mkannwischer]

Thanks! Now A55 performance looks good again, too.
LGTM.