Refactor collision objects and visual objects

[luator]

As discussed in #29, the collision and visual objects should be refactored and unified.

• Arguments for collision and visual shapes should be unified (right now one uses half extents and the other full extents). Maybe they can even be combined to avoid code duplication.
• Consistently use np.array instead of list everywhere for things like positions and orientations.
• Make sure all object types follow the same API (right now, there is for example visual_objects.Marker which has not been touched in Use small Cuboid instead of Cube #29.
• Add more functionality? (e.g. method to get velocity of a collision object)

Copy of my suggestion from #29:

The interfaces for collision and visual objects should really be unified, ideally even be merged in some way. I have not thought this through completely but maybe something like the following would be a good solution:

cube_shape = CubeShape(width=0.1)
collision_cube = make_collision_object(cube_shape, mass=5, position=...)
visual_cube = make_visual_object(cube_shape, position=...)

With this, the same "cube shape" object could be used for both the manipulation object and the goal marker, reducing the risk of accidentally defining them differently.

[luator]

@shrutij01 Does this make sense to you?

[shrutij01]

@shrutij01 Does this make sense to you?

@luator yes, it does! Thank you, I will look into this.