Update docs for predictions on UAV tiles

Two common issues in model predictions for UAV images:

1. Problem: The pixel values in tiles are very large integers/floats (e.g. >300). This is a common issue with orthomosaics. Solution: Simply divide all pixels by a large number (e.g. 10,000) for the model to predict polygons.
2. Problem: The ground sampling distance of drone images is usually quite high, while the model weights are for 20cm imagery. Solution: upsample your image to 20cm/pixel before running predictions.