Integrate C++ ranger in larger C++ program

[CopyOfA]

I am hoping to use C++ ranger in my C++ program, but the way the ranger codebase works now, I would need to initialize a forest model each time I want to use the model against different datasets. Moreover, the model expects the input data to be a file. What I am hoping to do is:

1. initialize a model from a .forest file, without data (I don't need any of the training parameters either, e.g., mtry, splitrule, sample_with_replacement)
2. pass an array into the model and collect model output
3. repeat 2 for an arbitrary number of arrays
4. free memory from model

The ranger codebase is extensive, so I'm wondering if someone has done this before, so I don't need to develop new methods within ranger. If not, I don't mind developing this, but I'll need some guidance on where best to make changes.

[mnwright]

Yes, implementing a proper library with a common API would be a great improvement. See also #304, #349, #420.

[CopyOfA]

In my application, I only need the classification forest, so I'll see what I can do on this. My plan is to write a derived class from ForestClassification and write all new initialization and data loading methods in that class, so I don't mess with the original code.