NumPy - fast-compact serializing (to bytes) arrays of popular dtypes

Question

I have an array a that may be of any popular dtype, including np.str_ or any standard Python type (stored as np.object_).