build: self-contained pip-installable Python wheel

[leonardoaraujosantos]

Summary

Closes the last Tier-1 adoption item: pip install ./bindings/python now builds and bundles the native C ABI, so import cyberfluids works with no separate manual C++ build.

How

• The binding switches to a scikit-build-core backend that drives the repo-root CMake build during pip install / pip wheel.
• A new CYBERFLUIDS_PYTHON_WHEEL CMake path:
  • forces the pinned FetchContent NumPP built static + PIC and links it into libcyberfluids_c → a single self-contained .so with no external libnumpp dependency;
  • links libtbb with a DT_NEEDED entry on Linux (--no-as-needed);
  • installs the library into the cyberfluids package via its own python install component, so the fetched NumPP's own install rules stay out of the wheel.
• For a wheel build, find_package(NumPP) is skipped so an installed shared NumPP under .deps can't break self-containment.
• __init__.py loads the bundled library from the installed package first (build-tree fallback and $CYBERFLUIDS_LIBRARY override retained).
• Adds bindings/python/README.md (PyPI long description).

Verification (clean venv)

pip install ./bindings/python        # compiles C ABI, fetches + static-links NumPP
python -c "import cyberfluids ..."    # from an arbitrary dir
# → version 0.0.1; loads .so from site-packages/cyberfluids/
# → 2D cavity  shape (48,48,2)  max|u|=0.0896
# → 3D tunnel  shape (48,24,24,3) max|u|=0.0629

ldd on the bundled .so: no libnumpp, no missing deps (only system libtbb/libstdc++). The 3 local python_* ctest failures remain the known Anaconda libstdc++ shadow (worked around here with LD_PRELOAD); CI's system Python is unaffected. Normal (non-wheel) build + all native tests unaffected.

Docs/spec

build-and-packaging spec gains a "self-contained Python wheel" requirement (+ scenarios); getting-started.md, roadmap.md, and CHANGELOG.md updated. openspec validate --all --strict → 14/14.

Follow-up (not in scope)

Publishing a manylinux wheel to PyPI (bundling libtbb/libstdc++ via auditwheel) is the remaining step for pip install cyberfluids from the index — tracked on the roadmap.