GitHub CI Workflow Enhancements Part 5

[ohearnk]

NOTE: do not merge this until the 3 gradient tests failures described below are fixed.
NOTE: the branch in the PR is based off of the work in #402, so merge that PR first before merging this work.

This PR adds support for building QUICK with the Intel oneAPI compilers (CMake via the -DCOMPILER=INTELLLVM keyword). Also Github CI tests for Intel oneAPI builds are added for the following runners:

CMake builds:

• Ubuntu 24.04 (x86_64): Intel oneAPI (v2024.2.1, v2025.0.1)

Notes / Limitations:

• Builds using the bundled BLAS/LAPACK code lead to a large number of test failures (~50%) but builds with Intel MKL (oneMKL) resolve these issues. If we want to use and test the bundled BLAS/LAPACK Fortran codes, this will need to be investigated and fixed (possibly by updating this code -- see, e.g., Update bundled BLAS/LAPACK code #393).
• This work backports the compilation flags used by AmberTools for Intel oneAPI builds (speficially, not using the -ipo Fortran flag due to related issues with interprocedural optimization for libraries).

Closes #327.

[ohearnk]

Notes on the 3 gradient tests failing (both serial & MPI versions):

• Last entry of point charge gradients is incorrect for each of tests ('XYZ' column)
• STO-3G basis set in common for tests
• Failures consistent across Intel oneAPI versions: 2024.0.2 (SDSC Chinotto), 2024.2.1 (Github), 2025.0.1 (Github)
• Valgrind reports no memory issues (SDSC Chinotto)
• No suspicious compilation warnings
  • One warning regarding variable name length exceeded in cshell gradient module (global namespace) but resolving this warning via shortening module names did not correct below test failures
• Example diffs from failing tests (serial version):
  • test/runs/serial/grad_naip_meoh_libxc_pbeh_sto3g.grad.dif:

704c704
<   193X     0.     0.
>   193X    27.9500174444     0.
705c705
<   193Y     0.     0.
>   193Y    27.9500174444     0.
706c706
<   193Z     0.     0.
>   193Z    27.9500174444     0.

• test/runs/serial/grad_rpsb_meoh_rhf_sto3g.grad.dif:

911c911
<   247X     0.     0.
>   247X    24.2576170724     0.
912c912
<   247Y     0.     0.
>   247Y    24.2576170724     0.
913c913
<   247Z     0.     0.
>   247Z    24.2576170724     0.

• test/runs/serial/grad_ncnaip_b3lyp_sto3g.grad.dif:

671c671
<   181X     0.     0.
>   181X    26.0000019855     0.
672c672
<   181Y     0.     0.
>   181Y    26.0000019855     0.
673c673
<   181Z     0.     0.
>   181Z    26.0000019855     0.

[ohearnk]

Failing test issues have been resolved, so this is now ready for review.

[agoetz]

Looks good to me. All serial and mpi tests of the full CPU test suite including f functions pass using oneAPI 2025.0.4.

We will have to address the limitations wrt bundled BLAS/LAPACK. At the very least document that cmake flags -DMKL_HOME=$MKLROOT -DMKL_MULTI_THREADED=FALSE are required.