Investigate OpenACC (offload) and OpenMP (host threading) issue when launched from Python

[pramodk]

Describe the issue

OpenACC (offload) and OpenMP (host threading) seems to interact badly when launched from Python. Typical error:

call to cuLaunchKernel returned error 400: Invalid handle

To Reproduce

TODO

See .gitlab-ci.yaml where comment is:

 # Shared + OpenACC + OpenMP host threading has problems
 build:coreneuron:mod2c:nvhpc:acc:shared:
   extends: [.build_coreneuron, .spack_nvhpc]
   variables:
     SPACK_PACKAGE_SPEC: +caliper+gpu~legacy-unit~nmodl~openmp+shared+tests~unified build_type=RelWithDebInfo

Expected behavior

Even if OpenMP is enabled on the host side, all tests should pass.

System (please complete the following information)

• OS/Platform: BB5
• Compiler: NVHPC 22.3 (or others as well I believe)
• Version: master
• Backend: GPU

Additional context

this issue is a result of a discussion /todos in #795