Four more pose methods that said they always return a figure

[alexarje]

Continuing the tail of #350. mypy 52 → 50, 682 tests pass.

The number understates this one. Annotating the six render_* helpers in _pose_visualize.py — which had no return annotations at all — first pushed the count up to 54, because four public methods in _pose.py declared -> MgFigure while receiving MgFigure | None from them.

So pose_waterfall(), pose_segments(), pose_center() and pose_distance() have always been able to return None — on too few frames, or an empty render — and have always said otherwise. Same shape as the nine audio methods corrected in 1.13.0: the code was right, the description was wrong. Signatures and docstrings now agree with it.

Two inference fixes in the same file:

• points.append(None) is deliberate, recording a keypoint below the confidence threshold as absent. The list was inferred list[tuple[int, int]] from its first append, so the intended None was flagged as the error.
• _trail is a deque or None depending on whether marker history was requested, unannotated at three sites.

Remaining after this: 50.

🤖 Generated with Claude Code