[FEA] Migrate .optimize() method to C++

[wphicks]

Currently, the logic for .optimize on FIL models is written in Python. Migrating it to C++ with a thin Python wrapper would allow it to be used by C++ consumers as well.

[wphicks]

@hcho3 For my own reference, was this closed as "will not fix" or is this implemented in C++ somewhere? I'm asking because it looks to me like the logic is still in the Cython layer, and I didn't immediately see a C++ implementation to complement it.

If this is a "will not fix," was there a specific reason? I'm asking because this is functionality that I might want in something that I'm working on and don't want to wander too far down that path if there is a good reason not to have it in the C++ layer. This also seems to have been of interest to @Kaiyang-Chen, and it might be useful in Triton (though I'm not sure what the project priorities are for that right now).

Apologies if I'm just missing the relevant new code!

[wphicks]

@hcho3 any further word on this? It's going to be relevant to work that I'm doing very soon, so I wanted to check in once more before launching into something that you may have already thought through and rejected.

[hcho3]

Reopening because it was closed by mistake.

We did not yet find time to work on the C++ implementation. If you'd like to work on it, I'd be happy to review your pull request.