Also exclude samples with Inf or NaN distance

[kawk]

After #18 the conversion of real world R2300 LaserScan to PointCloud sometimes fails, reporting malloc(): invalid size (unsorted). This update shall fix that.

[kawk]

The fix/nan-to-pointcloud is a quick fix but probably needs a bit more investigation as I'm not yet a point cloud expert. Now that invalid values always come as NaN or Inf, is range checking still needed? Does this check have to match some other implementation in TF code exactly?

[kawk]

The original code references https://github.com/ros-perception/laser_geometry/blob/indigo-devel/src/laser_geometry.cpp#L110

[kawk]

The loop in PointcloudPublisher::project_laser checks validity of points subtly different than laser_geometry::projectLaser_, mishandling NaN and Inf. Instead of acting on a point if it is inside the allowed range, it skips points where either (range<range_min)=true or (range>=range_max)=true ... but neither of these would be "true" with range==NaN because comparison to NaN usually yields false.

[kawk]

This fixes #24 - please pull.