Modify Conditional GAN to work with variable image sizes

[suryajasper]

Refer to following python script: https://github.com/suryajasper/SynthMining/blob/51f0f808d8dca6cf42f7eb808221438d7a15f926/synthesis_pipeline/condGAN/cond_DCGAN_network.py

      layers = []
      in_channels = channels + embedding_size

      for out_channels in conv_channels:
          layers.append(nn.Conv2d(
              in_channels=in_channels,
              out_channels=out_channels,
              kernel_size=4,
              stride=2,
              padding=1,
              bias=False
          ))
          layers.append(nn.LeakyReLU(0.2, inplace=True))
          in_channels = out_channels

      layers.append(nn.Conv2d(
          in_channels=in_channels,
          out_channels=1,
          kernel_size=4,
          stride=1,
          padding=0,
          bias=False
      ))
      layers.append(nn.Sigmoid())

      self.main = nn.Sequential(*layers)

1. Add optional keyword argument to Generator and Discriminator constructors called img_size. Assert that img_size is an integer power of 2
2. Rebuild sequential design of Generator to allow for variable img_size. Note that the sequential design features hidden convolutional layers, batch norms, and activations that reduce the channel size by half through the in_features / out_features and double the image size. These operations can be performed iteratively using a for loop that progressively increases the dimensionality of the modified latent vector into the image size.
3. Rebuild sequential design of Discriminator to allow for variable img_size. Note that the sequential design features hidden convolutional layers, batch norms, and activations that reduces both the channel size and image size by half. These operations can be performed iteratively using a for loop that progressively lowers the dimensionality of the input image into a single output node expressing the probability the image is real.