(alicevision#949).
For now, only the smallest family of tags (tag16h5) is supported,
which consists of 30 coded tags.
For each marker, 5 keypoints are generated:
- the center point (computed from the four corner points), with IDs 0-29
- the 4 corner points (in counter-clockwise order), with IDs 30-59, 60-89, 90-119, and 120-149
Thus, the feature descriptor has 150 dimensions.

The implementation follows along the lines of the CCTag markers, whereever possible.
In most places where CCTags were used before, AprilTags are now an additional option.
Exceptions are the camera and rig calibration and localization functions,
as well as a function which seems to create visualizations of feature matching results.

The Dockerfile for Centos has been adapted to start an AliceVision build using AprilTags.
I did not dare touch the Dockerfiles for Ubuntu, as the Centos one seemed to be the main one in use,
and it also was the one where the build process went quite smooth.

This has not been tested on Windows, sorry. It seems that AprilTags officially support only Linux,
although Windows should be possible according to the official repo.

There is one change which I do not really like, where it might make sense to change the
underlying classes/interfaces a bit to make different marker feature extractors easier.
This is the change in src/aliceVision/sfm/pipeline/ReconstructionEngine.cpp, which
only works by testing the results of dynamic casts on feature::Regions subclasses, although
the code that follows is rather similar for both CCTags and AprilTags, and could probably
substituted by something more general. The reason for this is that the marker ID is computed
from the descriptor, for which there could be an abstract function in the superclass.
I did not dare to make any such deep changes to the code, though.

- Removed all usages of OpenCV.
- Removed empty constructor definition in favor of "= default" in header.
- Changed scale to const.