fuse -> ROS 2 fuse_core: Fix Async (Redux)

[sloretz]

@methylDragon Here's an alternative to #293. I'm running the test_async_sensor_model test locally with --retest-until-fail 1000. I think this addresses the two issues found so far.

@svwilliams for visibility

[methylDragon]

I've updated the PR to include the following changes:

• Add an initialized_ flag that prevents an async class from getting initialized twice (with the appropriate tests)
• Change the tests to test with multithreaded executors (and made the init test more strict)
• Fix async issues by using reentrant callback groups on top of the chaining take_data()
• Update docstrings

We uncovered a potential bug in rclcpp's executors' handling of mutually exclusive callback groups, but in this case it doesn't block the progress for fuse_core

[methylDragon]

See: #276

[methylDragon]

Approving, but double checking with @sloretz that my changes are alright

[sloretz]

I made a few changes after self-reviewing and reading this bit of fuse_tutorials.

• In ROS 1 - start and stop are scheduling onStart and onStop and nothing else. I updated this PR to make sure the ROS 2 port behaves the same (meaning I removed the bits where we cleared the callback queue when the Context was still valid)
• There's an edge case in stop() that was fixed by Call onStop() directly if !ros::ok() in stop() #182. When ROS is shutdown the AsyncSpinner won't call anymore callbacks, so the onStop method won't get called. That's why there's a special case where onStop is called directly, and it also opportunistically took the opportunity to clean up the AsyncSpinner (though IIUC it doesn't have to do that). The ROS 2 executor has the same behavior of no longer servicing callbacks when the Context shuts down, so I made the logic the same as it was before.
• I fixed the race condition between initialize and destruction by making initialize() block until the spinner thread starts. This made internal_stop() simpler.

[sloretz]

@ros-pull-request-builder retest this please

Changes since review

[methylDragon]

I realized we should have initialized_ = false inside internal_stop so the user can choose to reinitialize the model

Aside from that, minor comments to consider

[methylDragon]

@ros-pull-request-builder retest this please

[methylDragon]

Let's add an initialized_ = false in the internal stops, other than that, LGTM!

[sloretz]

Let's add an initialized_ = false in the internal stops

Added in 9c95cf6

[sloretz]

@ros-pull-request-builder retest this please