Changes

• Fixed a bug in jpg dimension parsing. In get_jpeg_dimensions(...) we try to only read the first few bytes to obtain the correct width and height of a jpeg image. This failed for certain files as we treated a DHT table start as a marker which resulted in some non sense dimensions.
• We now have an explicit list of jpg SOF (start of frame) markers that we check for. If the given jpg file does not any of them we move on and use pillow to properly decode the image and then get the dimension.
• We added more unit tests to catch the issue.