Draft: core, network, internet, traffic control, routing: Drastically speed up simulations with large topologies and flows

Most likely will split per module. We had a lot of superlinear code being executed per packet. With large scale simulations (e.g. 100k flows) up to 47x speedup was observed (with now removed custom memory allocators to recycle buffers on highly frequently allocated objects, such as EventImpl, Packet, Buffer. They may come back in a follow-up iteration, since this MR is already too big without them).