pip build

[casperdcl]

• pip: use scikit-build-core to drive cmake
  • part of Conda install Google colab + Pip installation for CIL #932, PyPi Wheel #2097, upgrade build framework #1875
  • cherry-picked some commits from Build project with Scikit-build-core #2105 /CC @purepani
• update cilacc lib path
• remove Wrappers/Python/CMakeLists.txt
• update conda recipe
• update docker build
• update CI workflow
• update docs
• update CHANGELOG
• depends on cmake: FindIPP fallback #2148
• depends on CI: workflow_dispatch #2160

testing

1. cleanup any old installation (pip uninstall cil ; rm $CONDA_PREFIX/lib/*cilacc.*)
2. follow the updated README

open follow-up issues

• cilacc: static link intel-openmp,ipp #2154
• cilacc: fix openmp linking #2155
• PyPI dist upgrade build framework #1875
• drop data submodule upgrade build framework #1875 (related: data env var #2162)

[purepani]

There are sort of 3 separate tasks that need to be done in some order:

• switch to nanobind or pybind11 or similar
• simplify IPP library discovery
• switch to scikit-build

I was attacking it from the perspective that it would be easier to do 1 and 2 before 3(and I probably unnecessarily mixed up CI changes into there). This PR attacks it by doing 2 and 3 together.
How hard would it be to just simplify IPP library discovery in 1 PR? That seems like the simplest unit to change here, at least to me.

[purepani]

Oh I just realized - are you not doing the build with scikit-build-core yet?
The reason I changed the name for the folder was because I believe scikit-build-core needs the last component of the file name to be the same as the library(otherwise the build doesn't work).

Perhaps making that tiny change in a separate PR in anticipation is a good idea to minimize conflict issues.

Edit: OH I see what's happening; the python package is being built by cmake itself, whereas I just had the c code being built by cmake, which is needed since you're not building a c-extension but a normal shared library

[purepani]

After attempting it myself, it might just be easier to do all 3 at once(at least for me). I don't personally have the cmake knowledge to do the steps separately, and using scikit-build-core makes it way simpler.

[casperdcl]

Thanks @purepani.

simplify IPP library discovery in 1 PR

Good point; it's now in #2148

And yes, separate PRs make it easier for review/debugging/approvals.

[paskino]

currently trying to test conda build but failing, potentially because my conda is old-ish?

[gfardell]

All seems happy on windows