fix segfault at robot-model update

[dorezyuk]

PR fixes issue #245

The fix moves the ownership of the footprint from the optimizers to the teb_config class. While this change is debatable (keeping stuff separate is cool) it allows a straight forward fix of the tagged issue and makes dedicated interfaces for setting the footprint obsolete. It might also be acceptable to view the footprint as part of the config for a robot controller (but as said, also fine not to do so).

The positive side effects of this fix are

• the footprint is then constant for the optimal_planner and homotopy_class_planner
• the "default footprint" (PointRobotFootprint) is defined only once.
• one could make the footprint unique_ptr instead of shared, making the ownership clearer

[amakarow]

The positive side effects are clearly nice. However, as you already pointed out, I am still not hundert percent sure if we should really move the robot model to the config class ;).

[corot]

Equivalent change applied on #393