Change pointcloud_preprocessor components QoS policy

[VRichardJP]

As pointed in #333, the current default QoS policy for pointcloud_preprocessor components may cause important delays in the lidar processing pipeline.

The current queue size is set to 5, so when a "slow" component (e.g. voxel_grid_downsample_filter runs in 150~200ms on my machine) consumes messages slower than they are produced (my lidar sensor is running at 10Hz), the queue ends up being filled with old messages and the latest message is only processed after a long delay. So in my case, the localization and perception modules ends up working on data which is sometimes more than 1 second old. But even with a faster machine, the modules could still work on old data.

I don't see a case where a queue of 5 is actually useful: if the processing speed is high enough everywhere in the pipeline, no more than 1 message is ever stored in the queue; if the processing speed is not high enough, old messages accumulate in the queue while they should rather be dropped?

Should queue size be changed to 1 to avoid this issue?