Drop the pip-installed cmake from the pytorch base image

[rcannood]

This PR makes cmake usable again in base_pytorch_nvidia.

The nvcr image runs pip install cmake, which puts cmake, cpack and ctest shims in
/usr/local/bin. Those shadow apt's /usr/bin/cmake, because /usr/local/bin comes first
on PATH. The shims themselves look fine -- cmake --version works -- but they break as
soon as a build calls them from inside a pip build isolation environment, where
site-packages is no longer on sys.path:

File "/usr/local/bin/cmake", line 3, in <module>
    from cmake import cmake
ModuleNotFoundError: No module named 'cmake'

Reproduced on the published image, where it stops cellplm from building louvain:

docker run --rm --entrypoint bash openproblems/base_pytorch_nvidia:1 -c 'pip install cellplm'
# ERROR: Failed building wheel for louvain

Changes:

• Install cmake via apt and pip uninstall -y cmake. Uninstalling the package rather than
  removing the one shim also clears cpack and ctest, which would otherwise be left
  behind pointing at nothing.
• Add test_cmake.py, checking that all three resolve to the apt-provided binaries.

viash test passes 3/3, and pip install cellplm now gets through louvain-0.7.1. With
this in, the workaround in the cellplm component (apt: [cmake] plus
rm -f /usr/local/bin/cmake) can go.

🤖 Generated with Claude Code