Since each control point will call this when the path updates, the previous would correct the start segment 3 times instead of just once.
This fixes that.
Here we produce a local bound copy of the path version, and bind it to update the path type.
This way, if the path version updates (i.e. any control point changes type or position), we check that all control points have a well-defined path.
Additionally, if the control point piece is disposed of, the GB should also swoop up the subscription because of the local bound copy.
There was a bug where if you created a slider, moved the last point, and then added a point such that it became a PerfectCurve, it would fail to recover after becoming a Bezier.
This fixes that.
The `OnComplete` event was never being run due to the transform playing
out longer than the spinner's lifetime. I've matched the durations, but
also moved the `Stop()` call to what I deem a safer place to run it (I
did notice that without this it would still potentially never fire).
Note that this is more noticeable in the editor because of lifetime
extension. In gameplay, the returning of a spinner to the pool will
clean things up (but in the editor that can take longer, depending on
timeline zoom level).
Another thing worth mentioning is that the fade doesn't actually work.
This is due to https://github.com/ppy/osu-framework/pull/4212.
Closes#12119.
`SliderSelectionBlueprint.OnDeselected()` would expire the
`ControlPointVisualiser` on deselection, leading to its removal from the
blueprint and eventual disposal, but still kept a separate reference to
said visualiser in another field.
This could lead to that stale reference to a disposed child getting read
in `ReceivePositionalInputAt()`, crashing quite a ways down over at the
framework side on futilely trying to compute the bounding box of a
drawable with no parent.