Add Doppelgangers++ support

I’m interested in adding Doppelgangers++ as an optional pair-filtering step before mapping.

I tried exporting it to ONNX, following the direction used for other learned models in COLMAP. The export works, but the model is still over 1 GB in FP16, so bundling it directly may not be ideal.

Possible options are:

direct ONNX integration; a separate command that filters pairs in the COLMAP database; keeping the model external and exposing a lightweight interface from COLMAP.

Before implementing this further, I’d like to get feedback on which direction would be preferred, especially given the model size.

Related:

#2973 — learned model integration #4303 — ONNX Runtime support #1437, #2737 - Conditions that doppelgangers++ might be useful

Happy to work on the implementation if there is interest.