Test or remove BMI code from AVX2 backend

[hanno-becker]

The AVX2 rejection sampling code inherited from the official Kyber implementation some ifdef's regarding the availability of bit manipulation instructions. Those are untested at present.

Remove all code guarded by BMI, or (a) test it, (b) evaluate the range of targets where this is available, and (c) show that it actually helps performance (if it does).

For now I am inclined to just remove it. We cannot have untested code in the library.

Platform:
Which platform does this concern?

• platform independent
• aarch64
• x86_64
• rv64
• other

[hanno-becker]

@dkostic Can you comment on this by chance?

[dkostic]

Hi Hanno, I agree that it'd be best to remove the code. Although the flag says BMI, the code actually uses BMI2 instructions. Since Haswell, Intel has both BMI and BMI2; but AMD has processors that support BMI but don't support BMI2 (see here for details https://en.wikipedia.org/wiki/X86_Bit_manipulation_instruction_set).

[hanno-becker]

@mkannwischer Do you have time to take care of this?