Muscle parsing should not be a part of robot parsing

[Samir Menon]

It's leading to clunky and hard to understand code. Moreover, it creates weird errors if you're testing a muscle model without the muscles etc.

The primary reason for the joint parsing was to simplify graphics init.

Now that we're moving to a distributed visualizer that operates over redis, we can remove the clunkly dual init. This gives us cleaner code when it comes to the graphics setup.

[Samir Menon]

On second thought, and after looking at the code, it's clunky because muscles are partially parsed.

They're parsed in the robot parsed data. But not directly initialized in other things that use robot parsed data to initialize themselves. This is weird.

For instance, robot io data does not add muscles to its actuator set. This was the primarily clunk that needed to be removed.

Taking a step back, resolving the clunk requires scl-wide dynamic typing access. As such, resolving this clunk properly requires changes to the dynamic typing code. Moving that to a separate issue. Will return when that is settled.

[Samir Menon]

Moved the actuator set dyn init to rio (as it should have been). This is a general solution that will also work for other actuator types. It also preserves the clarity of the data structure setup.