Content-type hint for sharper reframe

Tell Katto what kind of video it is, talking head, presentation, gaming or cinematic, and the reframe stops guessing.

The guess is the hard part. A podcast with inserted footage and a sports broadcast look similar to a detector counting people in frame, and they need opposite framing: one should follow a face, the other should show the play.

The hint narrows the decision rather than forcing it. A talking head hint does not let the engine letterbox an entire clip, which is the failure it was added to prevent.