Move MyMPILib into NEO-2

[krystophny]

Summary

Move MyMPILib into NEO-2 and keep the PR scoped to that change.

What Changed

• add COMMON/MyMPILib as the in-repo home of the library
• move the MyMPILib user doc into DOC/UserDoc/MyMPILib.tex
• move the dedicated regression test into TEST/mympilib/
• wire COMMON against the local NEO2MyMPILib target instead of LIBNEO::MyMPILib
• restore HTTPS-based dependency fetches in cmake/Util.cmake and TEST/cmake/test_libneo_tag.sh
• remove the extra LIBNEO_SOURCE_DIR and FGSL_PATH wrapper/config plumbing from this PR
• add make test to the active PR workflow so the ported TEST/mympilib path runs in CI
• keep a narrow top-level Python interpreter hint so fetched libneo configures reliably with the system Python it still requires today

Why

MyMPILib is NEO-2-specific code. Keeping it in libneo made libneo carry an unnecessary MPI build surface and made ownership unclear.

Impact

• NEO-2 now owns its MPI abstraction directly
• the CI path keeps using HTTPS for dependency fetches
• the ported TEST/mympilib test is now exercised by the active PR workflow

Verification

$ make clean
rm -rf build

$ make
...
[438/438] Linking Fortran executable NEO-2-QL/neo_2_ql.x

$ make test
cmake --build --preset default
[1/2] cd /home/ert/code/NEO-2/COMMON/MyMPILib && /home/ert/code/NEO-2/COMMON/MyMPILib/Scripts/do_versioning.sh
Versioning MyMPILib...
[2/2] cd /home/ert/code/NEO-2/build/COMMON && /home/ert/code/NEO-2/COMMON/../ShellScripts/tag_version.sh
...
Test project /home/ert/code/NEO-2/build/TEST
    Start 1: nrutil_test
1/2 Test #1: nrutil_test ......................   Passed    0.07 sec
    Start 2: test_mympilib
2/2 Test #2: test_mympilib ....................   Passed    0.05 sec

100% tests passed, 0 tests failed out of 2

Related

• Companion libneo PR removes MyMPILib from libneo entirely.
• Related issue: #258

[GeorgGrassler]

@krystophny I took a view at it: The breaking of the golden record run comes most likely from the changed find_or_fetch function workflow. Now it requires ssh authentification in the ci workflow, whereas previously it was find with https. In general, I recommend not fixing everything in one single pr now and just focus on the main objective: moving MPILib into NEO-2. So my review:

• avoid drive-by fixes for now if not needed to do the pr
• only move MPILib to NEO-2 and fix the local build process to as was already done in the pr with add_subdirectory etc.
• add a test: the golden record test only runs in serial mode. It offers therefore no protection of the correct working of MPILib.
• add the ported over tests in TEST to the CI pipeline

I looked at MPILib in libneo myself again and it is already quite self contained. There should be no changes necessary at all. Ran a simple recursive diff and got the following back

differences.txt

[krystophny]

Addressed the review points in b53dc8c.

• Restored HTTPS-based fetch/clone paths in cmake/Util.cmake and TEST/cmake/test_libneo_tag.sh, so this PR no longer switches CI over to SSH auth.
• Removed the extra LIBNEO_SOURCE_DIR and FGSL_PATH wrapper/config plumbing again to keep the PR focused on moving MyMPILib into NEO-2.
• Kept only one narrow build fix: a top-level Python interpreter hint in CMakeLists.txt, because fetched libneo still configures Python today and CMake was otherwise selecting a broken user-local interpreter here.
• The ported TEST/mympilib test stays under TEST/, and I added make test to the active PR workflow so it is exercised in CI instead of only existing on disk.

Local verification:

make clean
make
make test

Result:

100% tests passed, 0 tests failed out of 2

[GeorgGrassler]

* Kept only one narrow build fix: a top-level Python interpreter hint in `CMakeLists.txt`, because fetched `libneo` still configures Python today and CMake was otherwise selecting a broken user-local interpreter here.

Where was it selecting a broken user-local interpreter? I could not find anything about that in the CMakeConfigs? Or was it doing that in libneo configure/build instructions? Cause then this should belong to the libneo fix and not on the consumer side.

@krystophny added also some more comments/questions directly add the changed code lines. That would need clarification first, before approval.

[krystophny]

Removed the hardcoded Python interpreter path in 10d7499. The root cause was a stale uv-managed Python 3.12 symlink in ~/.local/bin/ that lacked dev headers and NumPy — CMake picked it up before the system Python 3.14. Fixed on the local system by removing the broken symlink; the find_package(Python ...) call finds the correct interpreter on its own now.

Regarding the inline comments: I can see the reference to source-level comments but the GitHub API returns none — they may still be in pending/draft state. Could you submit the review so I can see and address them?