Implement Polyak averaging over a list of model checkpoints. Each checkpoint is represented as a dictionary mapping parameter names (strings) to numpy arrays. All checkpoints share the same set of keys and matching array shapes.
Your function should compute the element-wise arithmetic mean of each parameter across all checkpoints and return a single dictionary with the same keys, where each value is the averaged numpy array converted to a nested Python list via .tolist().
If the input list is empty, return an empty dictionary.