Add GPU backend selection with load_cuda(), load_metal(), load_amdgpu(), load_oneapi()

[anvika-singhal]

Enables GPU backend selection in komamripy, mirroring KomaMRI.jl's package extension system exactly. Users can now load CUDA, Metal, AMDGPU, or oneAPI backends to enable GPU acceleration with a single function call. Addresses JuliaHealth/KomaMRI.jl#736.

Included

src/komamripy/_backends.py (new)

• Julia: using CUDA -> Python: km.load_cuda()
• Julia: using Metal -> Python: km.load_metal()
• Julia: using AMDGPU -> Python: km.load_amdgpu()
• Julia: using oneAPI -> Python: km.load_oneapi()

Each function calls jl.seval() to load the corresponding Julia GPU package into the Julia session.

src/komamripy/__init__.py (modified)

• Imports and exports the four backend loading functions
• Updated docstring with GPU backend usage examples

examples/gpu_backend_selection_example.py (new)

• Demonstrates optional GPU backend loading
• Shows that simulations run on CPU by default (no backend required)
• Includes commented examples for each backend
• Automatically tested by test_integration_examples.py

Details

No Python-side state management needed, Julia handles backend detection and defaulting.

[cncastillo]

Just minor comments.

Do you have some "evidence" that this works? maybe add a screenshot of a succesful run using your pc or ultron.

I would also add one line showing km.load_cuda() to the README.

[anvika-singhal]

Installed necessary backends on first run. No duplicate installations on second run.