Add macOS installation guide, pinned dependencies and CLI test script

[EnricoTalpa]

Summary

This PR contributes macOS-specific installation documentation, a pinned requirements file and a CLI test script for verifying the full dependency stack with Metashape's internal Python.

Changes

README.md

Added a complete macOS installation procedure (6 steps) below the existing upstream content:

1. Clone the repository
2. Copy scripts to Metashape's auto-load folder
3. Xcode SDK symlink workaround (required when only Command Line Tools are installed — Metashape's Python 3.12 was compiled against MacOSX11.1.sdk which no longer ships with CLT)
4. Install Python dependencies into Metashape's isolated user-packages-py312 environment
5. Patch detect_objects.py to replace CUDA-only torch wheels with CPU-compatible ones
6. Verification steps (terminal test + GUI menu item)

Includes a compatibility table and installed file structure map.

mac_requirements.txt

Pinned requirements file for macOS Intel covering all scripts except detect_objects.py:

• numpy==1.26.4, onnxruntime==1.18.1, opencv-python==4.10.0.84, Pillow==10.3.0
• open3d==0.19.0, scipy==1.12.0 and their full pinned dependency set
• rasterio==1.4.3 (bundles GDAL 3.9.3, no system GDAL required)

test_libraries.py

CLI test script (13 tests) runnable with Metashape's internal Python without launching the GUI:

PYTHONUSERBASE="$HOME/Library/Application Support/Agisoft/Metashape Pro/user-packages-py312" \
/Applications/MetashapePro.app/Contents/Frameworks/Python.framework/Versions/3.12/bin/python3.12 \
test_libraries.py

Covers: numpy, Pillow, opencv, scipy, onnxruntime, rasterio, open3d, torch, torchvision, albumentations, pytorch-lightning, deepforest, faster-coco-eval.

Tested on

• macOS · Intel Core i7 (x86_64)
• Metashape Professional 2.3.1 (build 22580)
• Python 3.12.12 (Metashape internal)
• All 13/13 tests pass

Notes on detect_objects.py and macOS

The upstream script pins torch==2.7.0+cu128 which does not exist for macOS (PyTorch dropped Intel macOS support after 2.2.x). The README documents the required patch to torch==2.2.2 (CPU). The script's existing CPU fallback logic (line 552: "Using CPU (will be very slow)...") works correctly without any further changes.