Add sub-directories for external dependencies of tesseract_collision

[marip8]

This PR divides tesseract_collision into sub-directories based on the optional external dependencies each sub-directory introduces (bullet, FCL, VHACD).

Note: the function createConvexHull in the library tesseract_collision_core currently depends on bullet API, so I left the function signature in core but moved the implementation to the tesseract_collision_bullet library. This means that any code that wants to use createConvexHull (i.e. functions in tesseract_urdf) needs to link against the tesseract_collision_bullet library until we find a better solution for uncoupling the creation of convex hulls from bullet.

[Levi-Armstrong]

@marip8 Do you mind if we leave the createConvexHull in core and keep the bullet dependency for now.

[Levi-Armstrong]

Then it may be better to update this function to leverage qhull to break the dependency on bullet for the core.

[marip8]

Do you mind if we leave the createConvexHull in core and keep the bullet dependency for now.

To be clear, the function is still defined in core/common.h so it doesn't break API in other packages; only the definition of the function moved to the bullet sub-directory. The only change required for downstream packages is to link against tesseract_collision_bullet rather than tesseract_collision_core.

There is not an easy way for me to keep the definition in core because it would require all the same commands for finding and linking against bullet that I moved to the bullet subdirectory, which would defeat the purpose of this PR.

I can take a look at qhull to see if we can keep the current behavior without needing the linking hack.

[marip8]

qhull seems relatively difficult to integrate into tesseract_collision at the moment, so we decided to look into using boost to create convex hulls. It has an algorithm, but according to this issue it only works in 2D and not in 3D. I replicated the test case from this issue and was able to confirm that it does not in fact work correctly in 3D.

If we decide to go the qhull route, I would like to do it in a separate PR for more clarity on the changes. @Levi-Armstrong what do you think about this?

[codecov]

Codecov Report

Merging #641 (a84a870) into master (c8c5b29) will increase coverage by 1.24%.
The diff coverage is 96.15%.

@@            Coverage Diff             @@
##           master     #641      +/-   ##
==========================================
+ Coverage   87.13%   88.38%   +1.24%     
==========================================
  Files         162      178      +16     
  Lines        8786    10557    +1771     
==========================================
+ Hits         7656     9331    +1675     
- Misses       1130     1226      +96     

Impacted Files	Coverage Δ
...sseract_collision/bullet/bullet_cast_bvh_manager.h	100.00% <ø> (ø)
...ract_collision/bullet/bullet_cast_simple_manager.h	100.00% <ø> (ø)
...act_collision/bullet/bullet_discrete_bvh_manager.h	100.00% <ø> (ø)
..._collision/bullet/bullet_discrete_simple_manager.h	100.00% <ø> (ø)
.../include/tesseract_collision/bullet/bullet_utils.h	95.29% <ø> (ø)
...llision/bullet/tesseract_collision_configuration.h	100.00% <ø> (ø)
...act_collision/bullet/tesseract_gjk_pair_detector.h	75.00% <ø> (ø)
...t_collision/bullet/src/bullet_cast_bvh_manager.cpp	89.18% <ø> (ø)
...ollision/bullet/src/bullet_cast_simple_manager.cpp	89.90% <ø> (ø)
...llision/bullet/src/bullet_discrete_bvh_manager.cpp	92.80% <ø> (ø)
... and 55 more

[Levi-Armstrong]

How about just moving the createConvexHull to bullet directory. It should only require a few headers update and then the linking is not straight forward.

[Levi-Armstrong]

Also it looks like vhacd depends on bullet also which it is not linking against which is why windows build is failing.

[Levi-Armstrong]

Could also create a simple interface class like the ConvexDecomposition maybe ConvexComputer and then create an implementation in the bullet folder.

[Levi-Armstrong]

Minor fixes to this in PR #649