fix: skip aspect_ratio for kling-o1 models and detect image MIME via Pillow

kling-o1-pro and kling-o1-std silently fail when aspect_ratio is included
in the payload — they derive it from the input image. Added
VIDEO_ASPECT_RATIO_MODELS whitelist so only kling-elements and minimax-hailuo
receive the parameter.

Also switched image_to_base64 to use Pillow for format detection instead of
trusting the file extension, which correctly handles files saved with the
wrong extension.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

freepik_cli/api/models.py

@@ -107,7 +107,14 @@ VIDEO_IMAGE_FIELDS: dict[VideoModel, str] = {
     VideoModel.MINIMAX_HAILUO: "image",
 }
 
-# Kling Elements requires named aspect ratio slugs
+# Models that support an aspect_ratio parameter at all
+VIDEO_ASPECT_RATIO_MODELS: set[VideoModel] = {
+    VideoModel.KLING_ELEMENTS_PRO,
+    VideoModel.KLING_ELEMENTS_STD,
+    VideoModel.MINIMAX_HAILUO,
+}
+
+# Of those, these require named slug aspect ratios instead of "W:H" strings
 VIDEO_SLUG_ASPECT_RATIO_MODELS: set[VideoModel] = {
     VideoModel.KLING_ELEMENTS_PRO,
     VideoModel.KLING_ELEMENTS_STD,