Summary

• Adds a cross-platform one-click dependency installer adapted from the GeoAI QGIS plugin
• Downloads standalone Python + uv package installer, creates an isolated venv, and installs all required AI packages automatically
• Supports NVIDIA CUDA GPU detection with automatic CPU fallback
• Includes sam3 and its undeclared dependencies (psutil, scikit-image, scikit-learn) for Meta SAM3 support

New files

• core/python_manager.py — Downloads standalone Python matching QGIS version from astral-sh
• core/uv_manager.py — Downloads uv binary (pinned v0.10.6) per platform
• core/venv_manager.py — Orchestrates venv creation and package installation (4-phase pipeline)
• deps_install_dialog.py — QDockWidget UI with GPU detection, progress bar, install/cancel/reinstall buttons
• workers/deps_install_worker.py — Background QThread wrapper for non-blocking installation

Modified files

• _samgeo_lib.py — Added venv-aware path setup so get_samgeo() finds packages in the managed venv
• samgeo_plugin.py — Gates plugin launch on dependency check; shows installer dock when deps are missing
• metadata.txt — Bump version to 1.2.0

How it works

1. User clicks SamGeo toolbar → plugin checks if dependencies are installed
2. If missing, installer dock appears with GPU info and "Install Dependencies" button
3. Installation runs in background: Python (0-10%) → uv (10-13%) → venv (13-18%) → packages (18-95%) → verify (95-100%)
4. On success, installer dock closes and main SamGeo panel opens automatically
5. On subsequent launches, cached venv is detected instantly (~no delay)

Test plan

• Clean install: remove ~/.qgis_samgeo/, open QGIS, click SamGeo → installer dock appears
• Install flow: click Install → progress bar advances → success → main dock opens
• Model loading: select Meta SAM3, click Load Model → model loads
• Repeat launch: close/reopen SamGeo → skips installer, goes straight to main dock
• Cancel test: start install, click Cancel → stops, install button reappears