[C++][Parquet] Better simd unpack algorithm

[AntoinePrv]

Describe the enhancement requested

The current algorithm unpack algorithm in the bpacking files do not fully leverage SIMD operations and perform poorly on SSE4.2 (in fact they are not built for that reason).
They are also not available for multiple sizes.

We can take inspiration from

• http://arxiv.org/abs/1209.2137
• https://github.com/fast-pack/LittleIntPacker/blob/master/src/horizontalpacking32.c

Component(s)

C++, Parquet