[photon-targeting] Fix JNI loading

[Gold856]

This is still a hack, but it should work better than the old one.

For simulation:

• Now successfully passes smoketest for me!
  

For RoboRIO:

• photontargeting-java ships the linux .so's, because reasons
• Since photontargeting-jni is a JNI dependency in our vendor JSON, it will be extracted and deployed to the roborio as part of the deploy process to a directory on the java library search path (?)
  • Which we then have RuntimeLoader System.load for us as part of PhotonTargetingJniLoader

[mcm001]

Would like to see this tested on a robot <3 but ty!

[person4268]

I needed to add duplicatesStrategy = 'exclude' to the cppHeadersZip task to get this to build. Not sure if that's related to the PR or not (edit: it isnt). Will test PR on real robot this wednesday.

[mcm001]

Are you sure you don't have two PhotonVersion.h's floating around?

[person4268]

I tried running the clean task between builds. This is an otherwise unmodified worktree, as far as I know.

[person4268]

There are in fact two PhotonVersion.h'es. One in ./photon-lib/src/main/native/include/PhotonVersion.h and one in ./photon-lib/src/generate/native/include/PhotonVersion.h. I think those are conflicting (since the cpp header zipping is likely flattening the src/main and src/generate distinction), but we obviously want the generated one to win so just excluding duplicates naively probably isn't the best strategy?

[mcm001]

Agreed. The clean task seems not to be sufficiently aggressive with its cleaning of generated files too (needs to be fixed). On current main we would want to delete the generated one and keep the in-source one.

[Gold856]

Opened #1564 for tracking

[mcm001]

I've updated your PR description

[mcm001]

Needs testing on a Rio; happy otherwise