Fixing pymupdf#2348

We were returning arbitrary image type codes when reading the image binary from the PDF, e.g. "flate", even though they never would correspond to a meaningful image file extension.
This fix catches type codes not corresponding to known image file extensions and cause such images to be converted to PNG images (via an intermediate Pixmap).

fitz/fitz.i

@@ -2863,7 +2863,10 @@ if len(pyliste) == 0 or min(pyliste) not in range(len(self)) or max(pyliste) not
                     res = NULL;
                     img = pdf_load_image(gctx, pdf, obj);
                     cbuf = fz_compressed_image_buffer(gctx, img);
-                    if (cbuf) {
+                    if (cbuf && cbuf->params.type != FZ_IMAGE_RAW && 
+                        cbuf->params.type != FZ_IMAGE_FLATE && 
+                        cbuf->params.type != FZ_IMAGE_LZW && 
+                        cbuf->params.type != FZ_IMAGE_RLD) {
                         img_type = cbuf->params.type;
                         ext = JM_image_extension(img_type);
                         res = cbuf->buffer;