Windows Compatibility and Smaller Tiles

[GlassesNerd]

For Windows Compatibilty

1. In hyper_dataloader.py, changed the file paths to use os.path.join() instead of directly using "/" so that windows file path is also compatible.
2. Code still works as original on Linux/MacOS.
3. Added instructions in methanemapper/README.md to create a junction in windows to link the dataset folder
4. Fixed some typos ("No create soft links" -> "Now create soft links") (linking command had wrong argument, "train_dataset" folder does not exist)
5. Added .bat equivalent of .sh files for convenient running on windows

For 128px*128px tiles

1. Added command line argument small_tiles to indicate whether to use 128px tiles or 256px tiles
2. 128px tiles are joined with np.concatenate() to recreate original 256px tiles
3. used npz compression in addition to splitting the tiles to further decrease dataset size
4. 2020 mhs dataset size reduced from 1.07TB to 228GB
5. Training time increases about 10% on SSD, 20% on HDD