Update ccglib integration
Allocate larger temporary buffers for transposed input data and weights since these may be padded. This fixes a crash when running the code for instance with 55 stations. Minor changes to bBeamFormerPipeline
are also made to make the output look a bit nicer.