* Fixing race condition in cell permute 2 : OpenACC  [performance optimization]
* Update Unit Test / Documentation

Co-authored-by: Tristan Carel <[email protected]>

…#795)

* coreneuron and mechanism library can be built as shared and it
  enables launching coreneuron on GPU via python
* update MOD2C and NMODL fixes to handle GLOBAL variables
      See BlueBrain/mod2c/pull/78
      See BlueBrain/nmodl/pull/904
* removed acc/openmp global annotations for celsius, pi and secondorder
  and they don't need to be copied on GPU
* Pass Memb_list* as an argument for all common prototypes in order
   to support global variables via argument
* free ml->instance if not empty
* add link to libscopmath in neuron as well
* nrn_ghk is now declared inline.
* homegrown present table to avoid dynamic loading + acc_deviceptr limitations
* use -gpu=nordc and make #pragma acc routine seq functions inline
* drop -lscopmath as its folded in elsewhere
* random123 header reorganisation
* try and cleanup CLI11 handling.
* try and consolidate build logic
* some CORENEURON_ -> CORENRN_ for consistency.
* export OpenACC flags to NEURON separately as well as part
     of the whole ... -lcoreneuron ... link line.
* libcoreneuron.so -> libcorenrnmech.so, try and fix static builds
* do not enable OpenMP in shared/OpenACC builds.
* add rpaths inside nrnivmodl-core.
* accept a private destructor function pointer from generated mechanisms
* drop ${TEST_EXEC_PREFIX} that was causing simple tests to be executed on many ranks.
* CORENEURON_GPU_DEBUG: add environment variable that enables cnrn_target_* debug messages.

fixes #141

Co-authored-by: Olli Lupton <[email protected]>