Use a thread to run ROS executor

[ahcorde]

No description provided.

[ahcorde]

@mjcarroll some ideas why this might be usefull

Lower callback latency: callbacks are dispatched immediately instead of waiting up to one 30 Hz tick (~33 ms). This is critical for time-sensitive data such as TF transforms and sensor messages. ║

Update loop never blocked: the Qt render/update loop runs at its full rate without being delayed by executor work. This keeps the UI responsive under load.

Higher throughput: all N messages are processed faster because the executor is not rate-limited to 30 Hz ticks.

[ahcorde]

I made in this branch a benchmark test comparing both solutions.

This is the result

	Old spin_some 30 Hz	Old spin_some 60 Hz	New thread (any rate)
Mean callback latency	~1 690 000 µ	~847 000 µs	~700–5 000 µs
Tick budget stolen	30% (10 ms / 33 ms)	60% (10 ms / 17 ms)	0%
Update loop blocked	yes	worse	never

[asymingt]

I like this change. Are you certain that it won't introduce any data races on variables accessed from within executor callbacks, now that they are accessed from a separate thread?