Help me install PyHydroGeophysX and verify it on this computer.
Repository: https://github.com/geohang/PyHydroGeophysX
Official instructions: https://geohang.github.io/PyHydroGeophysX/installation.html
Desktop guide: https://geohang.github.io/PyHydroGeophysX/agents/desktop_studio.html

1. Inspect my OS, Python executable/version, available environment managers,
   and existing NumPy, Qt, VTK and geophysical packages. Ask whether I want
   Desktop Studio, Python workflows, or both, and which methods I need.
   Do not assume Conda exists or infer all package origins from NumPy alone.

2. Read the current official installation instructions and package metadata.
   Prefer a dedicated environment with a supported Python version. If I ask
   to reuse an environment, inspect its binary dependencies first. Use its
   existing package manager for those dependencies; avoid duplicate pip/Conda
   Qt and VTK installations. Show the plan before changing the environment.

3. Choose a released package or source checkout explicitly. For a release,
   use python -m pip install with the appropriate extras from the official
   instructions. For source, locate an existing checkout or download the
   official repository into a new directory, then enter the directory that
   contains pyproject.toml before using an editable install. Do not overwrite
   an existing checkout. Every pip command must use the chosen interpreter.
   For updates, locate the checkout currently imported by that interpreter;
   do not create another clone or worktree unless I request it. Record:
     python -c "import sys, PyHydroGeophysX; print(sys.executable); print(PyHydroGeophysX.__file__)"
   An editable source folder and an environment's Scripts launcher are normally
   in different locations; that alone does not indicate a broken installation.

4. Start with the CPU setup unless GPU acceleration is needed and supported.
   For CUDA, inspect the NVIDIA GPU, driver, OS, Python and dependency support.
   nvidia-smi reports driver capability, not the installed CUDA toolkit.
   Select a compatible PyTorch/CuPy/cuDSS combination using current official
   instructions; do not assume any CUDA 12 driver supports every CUDA wheel.
   Never install competing CuPy variants. If torch.cuda.is_available() is
   False, inspect the wheel, driver and device availability before diagnosing
   the cause. Report an unsupported GPU configuration and offer the CPU path.

5. Verify imports, the installed version and required engines. Download only
   the small example files needed for a check, using a downloader suitable
   for this OS, into a new test directory. Run a short example and report
   finite results, any warnings and the engine that actually executed.
   If testing ADTLERT, use a compatible dataset without remote electrodes;
   report a fallback rather than claiming that GPU execution succeeded.
   For ADTLERT, also run the three isolated checks (runtime, single, time-lapse):
     python -m PyHydroGeophysX.inversion.adtlert_diagnostics --report gpu-check.json
   Wait for the command to finish and inspect every stage's exit code and JSON
   result. CUDA availability, a passing single-survey test, or Studio self-test
   alone does not verify GPU time-lapse. A traceback line alone does not prove
   that a process crashed; preserve the complete log and final exit status.
   If this command is unavailable in an older release, report time-lapse as
   unverified and use the matching release's documented test, or offer an update.
   On failure, retain gpu-check.json and report each capability separately.
   OpenMP conflicts can affect Intel or AMD CPUs. Do not set KMP_DUPLICATE_LIB_OK,
   delete/rename DLLs, disable line search, or patch NumPy to make a check pass.
   Do not modify source algorithms as part of installation unless I explicitly
   request a code fix. Offer the CPU engine and verify it independently;
   do not reinstall the same package combination repeatedly.
   After successful verification, save python -m pip freeze and, for Conda,
   conda env export. These record this tested machine; they are not a universal
   GPU compatibility guarantee. Re-run the checks after dependency updates.

6. For Desktop Studio, install the desktop extras and the engines needed for
   my methods, then run:
     python -m PyHydroGeophysX.qt_apps.launcher --self-test
   Launch the application and give me the exact command to reopen it.
   Report the environment name, installation path and any unverified features.

   For AQUAH Auto to report, keep provider/model/API key settings in the right
   assistant panel. The OpenAI default is gpt-5.6-luna with medium reasoning.
   Folder classification sends filenames and short previews to that provider;
   report whether a real API call was verified or only offline tests passed.
   RAG uses local reference text. If I request MCP integration, install the
   optional mcp extra and test the stdio server's list_workflows tool. Leave
   numerical recipe execution disabled unless I request --allow-run.
   Never claim that a generic DEM/XYZ file was used by an inversion merely
   because the classifier recognized it; verify the supported geometry adapter.

Do not delete environments, overwrite projects, or accept third-party Terms
of Service on my behalf. If a step needs my action, explain what to do.
Do not report success while required verification is failing.