Update ICP and depending libraries to fix long-running crashes

[jk-ethz]

This PR updates the ICP repository sources, fixing several issues with ICP, mainly crashes that occur when running it for a long timespan (> 1 h).
The crashes were very sporadic and unpredictable, making them hard to pinpoint and debug. ICP mostly ended up with the message

terminate called after throwing an instance of 'PointMatcher<float>::DataPoints::InvalidField'
  what():  Point cloud has 31872 points in features but 32256 points in descriptors
/icp_node died from signal 6
/icp_node left a core dump

This was particularly annoying since ICP not only had to be restarted, but also reinitialized with the current estimated pose of the robot. By that time, the state machine mostly had stopped working already since the TF tree was not connected anymore.

The PR also updates libpointmatcher which had other bugs fixed in upstream repositories not merged into the branch we were depending on, yet.

Related:

• #492
• #14