Add CI build jobs

See merge request !123

Strangely, $CMAKE_RUNNER_ARGS is badly dereferenced on windows powershell.
If this variable contains more than one argument, it behaves as if all
the args were not given to cmake.
Therefore, all the deps dirs are missing, and cmake configuration fails.
After many hours of investigation, we have not found why powershell/cmake
fails to correctly deal with this command.

A solution described in the following thread solves this, but is windows specific.
https://stackoverflow.com/questions/33994462/powershell-not-passing-arguments-to-cmake-properly

To keep easier and more readable CI script, the CMAKE_RUNNER_ARGS argument
is split into independent optional options that can be specified on the runner.
There are more args, but this easier to maintain and track errors.

As the PV plugin is a shared library, we also need to ensure that the core
SLAM lib (which may be static) is compatible with position independent code.

If the Paraview plugin needs to be compiled, we must find Paraview before PCL.
Otherwise, if PCL is built with pcl_visualization module and uses another VTK
than the one provided by Paraview, it may override the correct include
directories and libraries.

All the necessary VTK/PV dependencies are already added by the function add_paraview_plugin.
By linking to all VTK_LIBRARIES (which contains all VTK and PV libraries),
this adds way to much unneeded dependencies, such as Qt5.

[doc] Rename ENABLE_slam to LIDARVIEW_BUILD_SLAM

See merge request !126

[feat] Add criterion to add keyframe

See merge request !124

* A frame is now a keyframe if the maps do not contain enough points (a
threshold is added)

* A size computer (function Size()) is added to the rolling grid to fastly get the current maps number of points

Set initial maps and pose

See merge request !122

Fix and clean ROS/catkin dependencies, compilation and installation

See merge request !121

Adds a subscirber to 'set_slam_pose' topic of type geometry_msgs/PoseWithCovarianceStamped.
This allows to set SLAM pose from RViz using the 'set initial 2D pose',
or by publishing a full 3D pose using an other localization node.

- Allow to load initial keypoints maps at node startup
- Allow to freeze these initial maps to avoid updating them
- Rename maps_saving/pcd_format to maps/export_pcd_format

When ROS wrapping is built, catkin scans all directories,
and builds only those detected as ROS packages (with a CMakeLists.txt and a package.xml).

Therefore, the ROS wrapping build entry point is slam/ros_wrapping/lidar_slam/CMakeLists.txt.
But at the node defined in this CMakeLists needs LidarSlam lib, the compilation of this lib must be triggered.
To build LidarSlam lib, the entry point is slam/CMakeLists.txt.
Therefore, in ROS wrapping, we need to do an out-of-source add_subdirectory
by specifying an additional build directory.
This build directory may depend on ROS build system being used.

The previous directory did not work with `catkin build` from catkin-tools package.
Now, it is properly defined to be more robust and work with both `catkin_make` and `catkin build`.

Remove BUILD_SHARED_LIBS from cache entries

See merge request !120

This variable does not need to be a cache entry.
This variable existence is checked in normal variables, then in cache entries.
This allows to easily set this as a normal variable in a parent scope,
without needing to set it as cache entry, which would override all next calls to this variable.

When needed, this variable may be set locally to the desired value.

[feat] Add odometry and IMU constraints to optimization

See merge request !114

* The residuals belonged to the SLAM class
* It now belongs to each manager
* A getter is added to SensorManager
* The functions to compute residuals are renamed

* Sensor constraint functions are changed from bool to void

* The cost function ptr is tested to add constraint

* An abstract class SensorManager is created.
   * It is templated by the measurement type.
   * It contains a measurements vector and basic vector managers, a current idx measure, a weight, a timeoffset and a function to check
     if the sensor data can be integrated in the SLAM optimization.

* Two sensor manager classes : IMU/Wheel odometry with their specific residual creation functions.

* Add sensor offset time as slam member

* Add time offset to constraint functions

* Add time offset computation in vtkslam. This uses packet reception
time and first point timestamp.

* Remove hard coded time values