extend Image.resolve_image() to support PIL images and/or numpy arrays

[A-Schmid]

From my understanding, it is currently only possible to add image annotations based on an image file.
When creating the annotation dynamically (e.g. with OpenCV), we have to take the long route of saving the annotation first and opening it again by passing a file path or PIL ImageFile object. Especially for high resolution images, writing to disk takes a while so performance would be improved significantly by just using the image directly from memory.