Make RandomDownsampleFilter preserve field information

[VRichardJP]

Currently, the RandomDownsampleFilterComponent nodelet ignores the input cloud fields and produces only XYZ clouds. Whether the input cloud contains XYZIR or XYZIRADRT data, the output is always XYZ. As a consequence, the filter can only be used at the end of the pointcloud preprocessing pipeline, while it would rather be suitable at the beginning.

For example, if the sensor I use produces too many points for my processing capabilities, I would want to change the preprocessing pipeline like this:

1. convert velodyne -> pointcloud
2. downsampling
3. rectify, filter, concatenate...

But if the downsampling does not preserve XYZIR or XYZIRADRT information, the filters following after may not work.

For others downsampling algorithms, I understand it could be difficult to preserve all the fields, but for a random downsampling I don't think it would cause any issue.