In the first video, the champion model is leaning forward while running. This does not mean the hitbox should be leaning forward as well. The screenshot you posted was not the closest that the blades came to your shadow, it practically was touching it as it first returned.
The thresh example is more obvious and easier to spot. Look at the shadows of the two things (champion and the hook). Just because the character model is higher in elevation and doing some sort of yoga pose doesn't mean that the hitbox is also going to float in the air and distort. It's going to move, yes, but it's all in a 2d environment and static in shape. Once those two hitboxes collide, threshes hook model snaps to the center of the enemy champion model for aesthetic purposes. Blitz hook is the same way.
The animation of the model doesn't have a bearing on what is actually happening for collision or ability interactions. Heck, the champion suffering in all of those videos has their own way of abusing that themselves. Caitlyn's E->Q combo wouldn't fire from the void if things were based off the model location instead of where she started the ability.