1. Update DataLoader to include cloud masks we create

[mcecil]

No description provided.

[mcecil]

add new class similar to HLSDataset to read in mask data. Then duplicate mask for each band.

• remove normalization
• remove random cropping and cloud masknig (both for our new HLSDataset and MaskDataset)

[mcecil]

I've added changes to create a MaskDataset object.

• Normalization appears to be done regardless of what is specified. There is a line of code in getitem of HLSDataset

channels = (channels - self.mean) / self.std

• this line is run regardless of how normalization is set. I've commented this out for the MaskDataset object. I am confused how this works for HLSDataset. Are we normalizing twice?
• I've also commented code for rescaling reflectance values, which we don't want for the mask.
• Do we need to pair images between the imagery and mask datasets? I am guessing yes. Train and val chips are defined in json file. If we treat the HLS and Mask datasets as separate objects, we are assuming that we have the same .tif files listed for HLS and mask datasets. In other words, the same # of chips, in the same order.
• Alternatively, we could create one HLS + Mask imagery object that contains 5 bands (4 HLS + mask), and then we can have functions to slice off the HLS part and the Mask part. I think this could be cleaner for keeping track of images.
  @HamedAlemo

[mcecil]

I still need to duplicate mask for each band.

[mcecil]

I added mask duplication. Currently hard-coded to repeat 4 times. We should decide if we want to keep separate HLS and Mask objects or make 1 combined object.