Single-particle diffusional fingerprinting: A machine-learning framework for quantitative analysis of heterogeneous diffusion