[avx2] add _mm_256_cvtepu{8,16,32}_epi{16,32,64}

[sifkarov]

[avx2]

• _mm256_cvtepu16_epi32
• _mm256_cvtepu16_epi64
• _mm256_cvtepu32_epi64
• _mm256_cvtepu8_epi16
• _mm256_cvtepu8_epi32
• _mm256_cvtepu8_epi64

An intermediate u8x4 type is needed for the cast in _mm256_cvtepu8_epi64, so I defined one inline just like it's signed counterpart. I see there might be some overlap with other open PR's (regarding the u8x4), but this seemed like the best route to take for now given the context and the desire to define these just like they are in avx2intrin.h.

[gnzlbg]

@GovernorTarkin thank you for the pull-request!

An intermediate u8x4 type is needed for the cast in _mm256_cvtepu8_epi64, so I defined one inline just like it's signed counterpart.

The i8x4 type is already available in the v32 module, I think I would prefer for the u8x4 type to be there as well.

[sifkarov]

@gnzlbg I must have overlooked the v32 module. In that case, I can also prune the existing duplicate i8x4 definition.

In the mean time, it looks like #183 adds the u8x4 that we'll need here, so I guess that concern is addressed once #183 is merged.

[gnzlbg]

I think that this PR is going to end up being merged sooner than #183 (the rest of the PR LGTM) so I would suggest fixing it here, we can amend #183 afterwards.

[gnzlbg]

Thanks.