Update photoscan reconstruction to use Meshroom 2025

[sadhana-r]

Closes #427

Updates OpenLIFU's photoscan reconstruction pipeline to be compatible with Meshroom 2025.1.0, replacing the previously supported Meshroom 2023.3.0.

Main changes:

Pipeline files (src/openlifu/nav/meshroom_pipelines/*.mg)

• Bumped releaseVersion from 2023.3.0 to 2025.1.0 and fileVersion from 1.1 to 2.0 across all five pipeline templates
• Updated node versions to match Meshroom 2025 (e.g. CameraInit 9.0 → 12.0)
• Added explicit describerTypes: ["dspsift"] to FeatureExtraction nodes, reflecting a new required field in Meshroom 2025
• Reformatted input references to match the Meshroom 2025 .mg file schema (fileVersion 2.0)

photoscan.py

• Replaced --cache CLI flag with --save <project_file> since Meshroom 2025 seems to ignore the --cache spec and instead writes cache output to MeshroomCache/ inside the working directory
• Updated cache_dir path accordingly (temp_dir / "MeshroomCache")
• Fixed node name reference from "CameraInit" to "CameraInit_1" to match the node naming convention used in the updated pipeline files

[sadhana-r]

I compared the output of Meshroom 2023 and 2025. The reconstruction looks accurate but I noticed there is a slight difference in the color appearance of the model. I haven't changed anything in the pipeline, so the texturing differences are likely due to changes in Meshroom itself - the release mentions some changes to color calibration.
Additionally, Meshroom 2025 now uses dsp-SIFT as the default for feature extraction, whereas Meshroom 2023 used standard SIFT features.

[ebrahimebrahim]

This looks great, thank you!

[ebrahimebrahim]

Thank you also for testing and comparing outputs