Kandinsky 3: Text-to-Image Synthesis for Multifunctional Generative Framework