nx_bench_hll_vs_cpc.nx

bench_hll_vs_cpc.nx -- head-to-head workload runner. Streams N distinct synthetic keys through OUR HLL_8 and OUR CPC at matched lg_k, then feeds estimates + memory into the comparator for a sealed verdict. This is sister-comparison (our vs our), validating Lang 2017's HIP claim that CPC beats HLL on accuracy at same memory. The harness extends naturally: swap CPC for a Python-datasketches stub to compare against the incumbent.