Towards End-to-End Structure Solutions from Information-Compromised Diffraction Data via Generative Deep Learning