Convert rct_optimizations to be a ROS-generic CMake package

[schornakj]

The first step at enabling #35. The work I had done previously converted the RCT packages to be ROS2 packages via Ament, which isn't an ideal strategy for us to use here, so this is a fresh take on the task.

• Remove Catkin dependencies, macros, and functions from rct_optimizations.
• Change minimum required CMake version from 2.8.3 to 3.5.0.
• Add a set of CMake macros (in rct_optimizations/cmake/rct_macros.cmake) to streamline package configuration and test setup. The idea is that the other RCT packages will be able to use these too. These macros are lightly modified from the ones used in Tesseract.
• Use a more "modern CMake" approach in rct_optimizations/CMakeLists.txt. It now provides its libraries through the rct::rct_optimizations target.
• Move test source code to a test directory, which is added as a subdirectory if the RCT_ENABLE_RUN_TESTING flag is True.
• Change how rct_image_tools, rct_ros_tools, and rct_examples find and link against rct_optimizations to accommodate the fact that it is no longer a ROS package.

TODOs:

• Questions for @marip8: Do the changes I made to the way tests are structured still allow CI to build and run tests? Do we have to add RCT_ENABLE_RUN_TESTING to .github_actions.yml?

[marip8]

Looks good overall. It didn't appear that any of the tests were run, and I think you are right about the CI config file needing an update. It looks like you can add CMake arguments to industrial-ci using the CMAKE_ARGS environment variable per its detailed documentation

A couple other comments:

1. We should target against Eigen per their documentation. I was hesitant to suggest this for Fix eigen build issue #39 since it required a CMake min version increase; since we're doing that in this PR, though, we should target against Eigen the "right" way
2. @Levi-Armstrong mentioned moving the Tesseract CMake macros into a standalone repository. I assume we replace the rct macros with those once their available. If not, would it make sense to put those macros in a different place now so they can also be used by rct_image_tools when the time comes to make it pure CMake?

[schornakj]

I've made some changes which I will push up once I get tests running through CI again:

• I added a rct_common package, which now contains CMake macros and the GTest external package.
• I changed the CMake arguments for building and running tests. Setting RCT_BUILD_TESTS=True will build the tests, and setting RCT_RUN_TESTS=True will cause them to be run using make test after they're built.

[schornakj]

CI currently fails with the error below, which I think is because the environment used in the Github Action doesn't have git installed.

Errors     << rct_common:cmake /root/target_ws/logs/rct_common/build.cmake.000.log
  CMake Error at /usr/share/cmake-3.5/Modules/ExternalProject.cmake:1757 (message):
  
error: could not find git for clone of GTest

[marip8]

That seems odd. Doesn't the Docker use git to clone the repository it's building as well as packages from .rosinstall files?

You can specify that additional packages be installed before running the build using the ADDITIONAL_DEBS environment variable with industrial_ci. Maybe try that with git?

[schornakj]

You're right, I didn't understand it quite correctly and git is already installed at the top level.

e: I think that the Docker container set up by industrial_ci doesn't have git installed. The source repositories are cloned before the container is initialized and then copied in afterwards.

[schornakj]

@marip8 I rebased onto the latest master branch and fixed the issue with linking to GTest on the xenial job.

I think that the rct_optimizations tests are being, with two caveats:

• Setting RCT_RUN_TESTS=True makes the tests run both when building and during the run_tests step, which makes the CI process take longer.
• The run_tests step doesn't appear to report that the rct_optimizations tests ran and succeeded.

rct_optimizations doesn't directly link against Eigen (which is instead brought in by Ceres), so I'll address your comment about the Eigen target in a separate (and imminent!) PR for the other packages.