What is deep copying of objects (copy.deepcopy) in Python from a performance perspective? How does this mechanism bypass cyclic references, why does it work critically slowly on large object graphs, and what alternative patterns replace it in high-load systems?