Vision Update to Main

[a-krawciw]

Nearly complete work-in-progress to bring the vision pipeline back online.

[a-krawciw]

Merging of real Visual Teach and Repeat back into main!

There are a few changes to the compilation workflow to support multiple pipelines.

We have decided to make sensor selection a compile-time decision. While it is possible to decide at run-time, this ensures that the code is optimized for the sensor in use, and prevents accidental subscription to multiple sensors at the same time.

Additionally, this reduces compilation time significantly because unused pipelines are skipped.

1. To select the pipeline at compile-time the new env variable VTR_PIPELINE has been added. Use the options LIDAR, RADAR, VISION, RADAR-LIDAR to select your preferred pipeline.
2. The default dockerfile uses the GPU with a full open-CV build. The OpenCV build is slow, but the hassle of removing the dependency is too high.
3. Config files exist for the common configurations of the robots. Testing was performed on the Warthog robot. Other robots will require new config files.