Add FAQ to Documentation

[adamjstewart]

Users coming from torchvision are going to encounter a lot of common issues:

• Used built-in sampler instead of GeoSampler
• Used built-in collate_fn instead of collate_dict

We should have an FAQ with the error message you get when you make one of these mistakes and instructions on how to fix it.

[adamjstewart]

Another thing worth adding is a list of best practices for speeding up sampler performance:

• Warp all files to the same CRS
• Resample all files to the same res
• Use TAP
• Use COGs
• Experiment with compression
• Experiment with different file formats
• Experiment with different dtypes
• Block size that is a multiple of patch size
• Largest patch size and batch size that fit in memory
• Experiment with GDAL_CACHEMAX
• Use settings from https://github.com/pangeo-data/cog-best-practices

This would have helped @yichiac