Issue while running patstim test on GPU from testcorenrn repository

[pramodk]

Same as #501:

Trying to run testpatstim.hoc test with CoreNEURON on GPU and see:

$ ./x86_64/special-core -e 100 -d coredat --gpu --mpi --pattern patstim.spk
 num_mpi=1
 num_omp_thread=2

 Info : 4 GPUs shared by 1 ranks per node

 Duke, Yale, and the BlueBrain Project -- Copyright 1984-2020
 Version : 0.21.0 f2e152b (2021-03-04 00:45:29 +0100)

 Additional mechanisms from files
 Gfluct3.mod exp2syn.mod expsyn.mod hh.mod hhderiv.mod hhkin.mod hhwatch.mod nacum.mod netstim.mod passive.mod pattern.mod stim.mod svclmp.mod vecevent.mod

 Memory (MBs) :             After mk_mech : Max 217.5625, Min 217.5625, Avg 217.5625
 Memory (MBs) :            After MPI_Init : Max 217.5625, Min 217.5625, Avg 217.5625
 Memory (MBs) :          Before nrn_setup : Max 218.5898, Min 218.5898, Avg 218.5898
 WARNING : GPU execution requires --cell-permute type 1 or 2. Setting it to 1.
 Setup Done   : 0.00 seconds
 Model size   : 10.96 kB
 Memory (MBs) :          After nrn_setup  : Max 226.6172, Min 226.6172, Avg 226.6172
GENERAL PARAMETERS
--mpi=true
--gpu=true
--dt=0.025
--tstop=100

GPU
--nwarp=0
--cell-permute=0

.......

 psolve |========================================================| t: 100.00 ETA: 0h00m01s

Solver Time : 1.45126
FATAL ERROR: variable in OpenACC API data update in routine was not found on the GPU

Everything looks fine but when GPU data is copied back to host then there seems to be an error. (basically update_nrnthreads_on_host() function)