General Identifiability and Achievability for Causal Representation Learning