DRAFT: fix undefined casting behavior through new SGRawValue type

[heshpdx]

We are seeing undefined behavior due to the casting of enums as integers, and that comes from calling the casted function pointers to tie property values.

This PR is a proof-of-concept of how to fix the issue using std::function with lambdas to wrap the casting. We should figure out if this is the best approach for changing the code. Right now this patch fixes the first error exposed by the LLVM optimizer, but there are 25+ more files of changes we would have to take on to make it clean.

#834

[heshpdx]

This line is causing the segfault in the TestPlanet CI test.

The issue has to do with BindingInstance in PropertyState being set to nullptr. In FGFDMExec.cpp line 796, a special 'Unbind()' assumes that all of the properties under 'atmosphere' have a BindingInstance set to that model. This is new code not in the SPEC CPU codebase.

[bcoconni]

Thanks for the PR.

I have a number of comments:

1. If we choose to address this issue by creating a new class inheriting from SGRawValue then it cannot be located in src/sim gear/props.[ch]xx for the reasons I have exposed in JSBSim for SPEC CPUv8 #834 (comment). Instead the new class should be located in src/input_output/FGPropertyManager.h so that JSBSim can compile in FlightGear.
2. Once we have decided to use a new class, using lambdas and std::function seem a bit overkill to me. I'd rather pass the enum value to the class constructor and it would be stored as a member of the class. That way the calls to FGPropertyManager::Tie would look very much like what they currently are.

[bcoconni]

Right now this patch fixes the first error exposed by the LLVM optimizer, but there are 25+ more files of changes we would have to take on to make it clean.

This is mostly addressed in the PR #1304 so that we can focus on the actual issue in this PR.