`./uw build` should verify numpy ↔ petsc4py ↔ Cython-extension ABI consistency (and rebuild on mismatch)

[lmoresi]

Problem

After the numpy≥2 dependency bump (#301/#304/#305) landed on development, import underworld3 fails on an environment whose petsc4py and compiled UW3 Cython extensions were built against numpy 1.x. The failures are cryptic and happen at import/first-use, not at build time:

# petsc4py (built against old numpy):
ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject
  ...  in init petsc4py.PETSc

# UW3 extension (built against old numpy), later at runtime:
ImportError: numpy.core.multiarray failed to import
  ... in init underworld3.function._dminterp_wrapper

Both the development main checkout and a worktree env hit this. ./uw build completed "successfully" yet left an unimportable install — its post-build import check aborted at petsc4py before recompiling, and pip install . did not rebuild the locally-built petsc4py binding.

What actually fixed it (manual, per-env)

1. Rebuild the petsc4py binding against the installed numpy:
   pip install --no-build-isolation --force-reinstall --no-deps $PETSC_DIR/src/binding/petsc4py
2. Clean-rebuild the UW3 extensions: rm -rf build/lib.* build/temp.* && ./uw build

Ask

This is an environment/build-consistency issue, not a source bug — so ./uw build (and/or ./uw doctor) should detect and handle it:

• Detect when the installed numpy differs from the numpy petsc4py / the compiled UW3 extensions were built against (e.g. compare recorded build-time numpy ABI, or a cheap import-probe of petsc4py.PETSc after build).
• On mismatch, rebuild petsc4py + force a clean extension rebuild, or at minimum fail loudly with an actionable message ("numpy X installed but petsc4py built against Y — run …") instead of leaving a broken install that errors cryptically at first use.
• The post-build import check should exercise a real path (e.g. uw.function.evaluate), not just import underworld3, so extension-ABI breaks surface at build time.

Environment

• numpy 2.4.6 (pin >=2,<2.5), PETSc 3.25 (petsc-325-uw-openmpi), locally-built petsc4py 3.25.0.