Tiled processing on MDv5 requests that are larger than 1280px wide

[nathanielrindlaub]

This is interesting & could be worth implementing–perhaps on the model endpoint side–but for images that are larger than MDv5's input (1280px wide), Dan has found that if you cut the image up into 1280px wide tiles and submit those individually for inference, then stitch the results back together, it can help improve detection of small animals (see Dan's note and tiling script for reference).