Fix GPU SIFT flag names for COLMAP >=3.13

[bjoernellens1]

Summary

COLMAP's #3465 (released in 3.13.0) renamed the use_gpu/gpu_index options from SiftExtraction/SiftMatching to FeatureExtraction/FeatureMatching, as part of generalizing feature extraction beyond SIFT. The old option names were dropped, not kept as aliases.

nerfstudio/process_data/colmap_utils.py still uses the old --SiftExtraction.use_gpu/--SiftMatching.use_gpu flags unconditionally. Against any COLMAP built from a recent enough source (>=3.13.0), feature_extractor/*_matcher reject these as unrecognized options, so --gpu silently has no effect (or the command errors, depending on COLMAP's exact CLI parsing behavior for unknown options).

Not specific to any particular GPU backend — this affects standard CUDA COLMAP builds too, once built from source at 3.13.0+. Found while validating a COLMAP fork built from a recent source tree, not something specific to that fork.

Fix

Version-gates the flag prefix using the colmap_version value this file already computes via get_colmap_version(), following the same pattern already used a few lines below for a different version-dependent option (if colmap_version >= Version("3.7")).

Test plan

• Verified the version-gating logic across the 3.13 boundary (3.8 -> SiftExtraction, 3.13 -> FeatureExtraction, 4.2.0.dev0 -> FeatureExtraction)
• Confirmed --FeatureExtraction.use_gpu/--FeatureMatching.use_gpu against a real COLMAP 4.2.0.dev0 build's feature_extractor -h/*_matcher -h output