Visualize eval results for frame V2 - use new annotation loader

Summary:
Use D32096988 to get predictions from file. Refactor out json loading logic from here.

I opted to leave bbox/kpt filtering logic here rather than in the loader, because I don't feel it's appropriate for a loader to modify/truncate/determine what it is loading.

---------

References:
* D31199879
* [visualizer.py](https://www.internalfb.com/code/fbsource/fbcode/portalai/detectron2/tools/inference/visualizer.py)
* [frame_grabber.py](https://www.internalfb.com/code/fbsource/[9d0cc082a413]/fbcode/portalai/data/core/frame_grabber.py)
* [Sample worldmodel prediction, from the json](https://www.internalfb.com/phabricator/paste/view/P462408906)

Reviewed By: abhinavgar

Differential Revision: D32119148

fbshipit-source-id: b10ce49cb647322bec9ab625cc5f3014c261425e

Author: Carolyn Soo

detectron2/utils/visualizer.py

@@ -800,6 +800,7 @@ def draw_and_connect_keypoints(self, keypoints):
         visible = {}
         keypoint_names = self.metadata.get("keypoint_names")
         for idx, keypoint in enumerate(keypoints):
+
             # draw keypoint
             x, y, prob = keypoint
             if prob > self.keypoint_threshold: