Lookup data structures not taking batching into account

[neworderofjamie]

I've been trying to use custom updates to do more of the resetting style stuff in mlGeNN and everything was breaking as batch size increased. Turns out this was totally broken as, for custom updates targetting neurons, the indices within which threads binary search to find their population were generated as if batch size = 1.

None of your models use custom updates on duplicated (i.e. batched) neuron group variables do they @tnowotny?

[codecov]

Codecov Report

Merging #507 (6ebdc0d) into master (a84bfa9) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #507      +/-   ##
==========================================
- Coverage   87.04%   87.03%   -0.01%     
==========================================
  Files          84       84              
  Lines       18104    18108       +4     
==========================================
+ Hits        15758    15761       +3     
- Misses       2346     2347       +1     

Impacted Files	Coverage Δ
include/genn/genn/code_generator/backendSIMT.h	96.51% <ø> (ø)
src/genn/backends/cuda/backend.cc	82.76% <100.00%> (ø)
src/genn/backends/opencl/backend.cc	91.59% <100.00%> (ø)
src/genn/genn/code_generator/backendSIMT.cc	96.06% <100.00%> (+0.01%)	⬆️
src/genn/backends/cuda/optimiser.cc	72.77% <0.00%> (-0.18%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a84bfa9...6ebdc0d. Read the comment docs.

[tnowotny]

Looks good - cannot say yet whether it makes a difference for eventprop. Potentially not (would have been too convenient)