Have a light scalar implementation

[fulmicoton]

The current implementations generate large binaries because they have one specialized implementation for each bitwidth, and do loop unrolling.

Add a flag-enabled implementation that uses a more compact scalar implementation. This would be useful for web assembly for instance.

[riemass]

Hello, could you elaborate a bit more about this task?
If I get it correctly, one simpler implementation should take care of bit packing. Should it cover all bit widths or have some generic parameter to specify the width?
Cargo config should in compile time decide which one to compile.

[fulmicoton]

All bit width from 0 to 32 included, but it should be a single method for all bit width (the bitwidth would come as an argument), and have no loop-unrolling.

Cargo config should in compile time decide which one to compile.

Yes