[FEA] Improve Perf of pack/unpack codepacker utilities for IVF-PQ

[tarang-jain]

Some code sizes (4 bits and 8 bits) pack without gaps into our IVF-PQ interleaved format. Packing and unpacking for them can be reduced to simple chunk copies (coalesced 128 bit reads), rather than reading (extracting) and writing each code one at a time.