Add 8/16 bits AVX2 swizzle

[AntoinePrv]

Add 8 and 16 bits AVX2 swizzle implementation.
These two sizes are not available on AVX.
Algorithm very similar to GH-1189.

[serge-sans-paille]

looking at the build failure on x86_64, you need to add some std::enable_if to disambiguate.

[AntoinePrv]

Or could it be that I should add the 8/16 bits overload as batch constants as well (which we'd also want)?

[AntoinePrv]

Seems like the failure is already on main.
The batch_constant is simply a naive fallback, I will address them in a separate PR.
CC @JohanMabille

[serge-sans-paille]

Looks good, squash everything with a nice commit message and we're good to go!

[serge-sans-paille]

Nit: could you use a variadic argument count here and just static_assert on the count? (here and the function below).

[AntoinePrv]

@serge-sans-paille you should be able to squash-merge directly from the Github UI once ready.
Let me know if you prefer I squash instead.

[serge-sans-paille]

@AntoinePrv : I've done it but I generally prefer when the author creates a consistent commit message during the squash: the author knows better!