Improve std::thread cleanup

[olupton]

• NEURON builds using NVHPC were failing after Replace pthreads with std::thread and friends #1859 because worker threads were not always joined before shutdown.
  • It seems that NEURON does not deallocate its data structures at shutdown.
  • Unclear why this problem was specific to NVHPC
• See Print memory footprint of each mechanism BlueBrain/CoreNeuron#833 (comment) for an example of a failing run.
• Reorganise code in multicore.cpp to join threads at shutdown before the std::thread objects are destroyed.
• Drop unused code with BENCH* macros
• Drop unused allocation function wrappers. std::malloc is specified to be thread-safe.
• The busywait branches have not been extensively tested, perhaps they can be dropped too?

[olupton]

https://bbpgitlab.epfl.ch/hpc/coreneuron/-/pipelines/62004 is testing this with the Intel and NVIDIA compilers.

[codecov-commenter]

Codecov Report

Merging #1880 (b50be5e) into master (68923c0) will decrease coverage by 0.00%.
The diff coverage is 84.40%.

@@            Coverage Diff             @@
##           master    #1880      +/-   ##
==========================================
- Coverage   47.13%   47.13%   -0.01%     
==========================================
  Files         543      543              
  Lines      112917   112900      -17     
==========================================
- Hits        53224    53213      -11     
+ Misses      59693    59687       -6     

Impacted Files	Coverage Δ
src/scopmath/sparse_thread.c	95.93% <ø> (-0.07%)	⬇️
test/unit_tests/unit_test.cpp	100.00% <ø> (ø)
src/nrnoc/multicore.cpp	88.81% <83.96%> (+0.48%)	⬆️
src/nrnoc/init.cpp	87.99% <100.00%> (ø)
src/parallel/ocbbs.cpp	65.21% <100.00%> (+0.08%)	⬆️
share/lib/python/neuron/rxd/species.py	78.85% <0.00%> (+0.07%)	⬆️

📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more

[azure-pipelines]

✔️ 9ca53e6 -> Azure artifacts URL

[olupton]

https://bbpgitlab.epfl.ch/hpc/coreneuron/-/pipelines/62004 is testing this with the Intel and NVIDIA compilers.

This looks fine.

[olupton]

✔️ 9ca53e6 -> Azure artifacts URL

Launched https://github.com/neuronsimulator/nrn-modeldb-ci/actions/runs/2556232205 comparing this PR to neuron-nightly

[azure-pipelines]

✔️ b50be5e -> Azure artifacts URL

[pramodk]

LGTM