Fix BINARY_ALIGN in a few OpenCL formats

[magnumripper]

It's amazing we still have this problem 12 years after introducing the alignment warnings.

[solardiz]

FWIW, as I recall from PHC list discussions, only char and unsigned char are allowed to alias any type. uint8_t isn't. Of course, in practice it's usually handled just the same as unsigned char, but theoretically a compiler is free to optimize things away when uint8_t * aliases something larger. So I suggest changing those two instances here to unsigned char *.

[solardiz]

It's amazing we still have this problem 12 years after introducing the alignment warnings.

I think the reasons for this are:

1. Just because greater expected alignment is not guaranteed does not mean it won't be there in a specific build/run - quite often it will be.
2. I think our CI runs don't check for those warnings at all - they're just buried in the logs. Maybe we should fix that.
3. Our CI setup tests only a handful of OpenCL formats (as part of ci/circleci: encoding-opencl, which is mostly an encoding test). We should add explicit OpenCL tests.

[magnumripper]

only char and unsigned char are allowed to alias any type. uint8_t isn't.

So the standard says "a character type". That begs the question whether wchar_t is "a character type" and consequently can alias anything. I expect the answer to that is "no" which makes me really hope OpenCL's uchar is "a character type".

I'll fix it, to increase the chance of remembering this.