Fix multiple issues with sparse synapse index narrowing

[neworderofjamie]

In #274 I implemented support for 'narrow' sparse synapse indices so, if the postsynaptic population was < 65536 neurons, uint16_t would be used and if < 256 neurons, uint8_t. However:

1. PyGeNN always tried to access the sparse synapse indices as uint32_t - it now callsSynapseGroup::getSparseIndType to get correct type
2. In the generated code for pushXXConnectivityToDevice and pullXXConnectivityToDevice, unsigned int was always being used - this also now uses the correct type.

[codecov]

Codecov Report

Merging #460 (c80ad12) into master (0cff2ed) will increase coverage by 0.02%.
The diff coverage is 87.50%.

@@            Coverage Diff             @@
##           master     #460      +/-   ##
==========================================
+ Coverage   87.23%   87.25%   +0.02%     
==========================================
  Files          78       78              
  Lines       17057    17056       -1     
==========================================
+ Hits        14879    14883       +4     
+ Misses       2178     2173       -5     

Impacted Files	Coverage Δ
src/genn/genn/synapseGroup.cc	82.74% <85.71%> (+0.85%)	⬆️
include/genn/genn/synapseGroup.h	90.90% <100.00%> (ø)
src/genn/genn/code_generator/generateRunner.cc	95.85% <100.00%> (ø)

---

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 0cff2ed...c80ad12. Read the comment docs.

[tnowotny]

Oh dear ... good catch there. I suspect this came out from the guy's report about huge post-indices fro a small post-population?
Can you tell that you have caught everything now (I unfortunately can't ...)?

[neworderofjamie]

It did indeed...The updated feature test definitely now covers the second bug but, PyGeNN is currently not well tested as most of the testing (aside from that done at a higher level by PyNNGeNN or mlGeNN) happens at the C++ level. For the low-level unit tests I think that is advantageous but, as part of the test revamp we've been talking about, maybe porting all the feature tests to PyGeNN is the solution.