fix bug in output format for pyav (#6672) (#6703)

* fix bug in output format for pyav

* add read from memory with constructor overload

* Revert "add read from memory with constructor overload"

This reverts commit 14cbbab239165be05096fd6cbb88cb0448502436.

* run ufmt

torchvision/io/video.py

@@ -273,15 +273,17 @@ def read_video(
         raise RuntimeError(f"File not found: {filename}")
 
     if get_video_backend() != "pyav":
-        return _video_opt._read_video(filename, start_pts, end_pts, pts_unit)
-
+        vframes, aframes, info = _video_opt._read_video(filename, start_pts, end_pts, pts_unit)
+    else:
         _check_av_available()
 
         if end_pts is None:
             end_pts = float("inf")
 
         if end_pts < start_pts:
-        raise ValueError(f"end_pts should be larger than start_pts, got start_pts={start_pts} and end_pts={end_pts}")
+            raise ValueError(
+                f"end_pts should be larger than start_pts, got start_pts={start_pts} and end_pts={end_pts}"
+            )
 
         info = {}
         video_frames = []