Release the weight for 4x RRDB architecture

Hello! We have retrained the model with the same hyperparameter setting for RRDB-6B (Which is the same structure as Real-ESRGAN Anime6B). In the future, when we have more time, we will tune more to the perpetual loss-related hyperparameter. More information will be found in the model zoo.

So why RRDB? This is because I can directly use my FAST inference codebase without modifying the code too much (lol).

Update: We have cleaned the 4x_APISR_RRDB_GAN_generator weight to have less weight size. Thanks!

Thanks!