Fix hashing bugs

[neworderofjamie]

When calculating the hashes used to merge groups together for initialization, usually all that we care about is the names and types of the state variables and how they are initialised but with SIMT backends where additional data structures are required for synapse dynamics/postsynaptic updates this is not enough. This fix simply includes whether weight update models include synapse dynamics and postsynaptic update code in the hash used to determine mergeability - this ignores the backend specific-details so will sometimes split initialization unnecessarily on CPU but fixes this (nasty) issue.

Furthermore, while trying to write a unit test to test this, I found a subtler issue that could also potentially cause issues. Essentially hashing an empty container (std::vector, std::string etc) had no effect on the hash so the following code:

Utils::updateHash("", hash);
Utils::updateHash("test", hash);

generates the same hash as:

Utils::updateHash("test", hash);

I've fixed this by hashing the size of the container before hashing the contents (which, one way or another, involves looping through the bytes of the contents). This PR fixes both issues.

[codecov]

Codecov Report

Merging #508 (fa1c9f2) into master (09b74db) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #508      +/-   ##
==========================================
- Coverage   87.03%   87.03%   -0.01%     
==========================================
  Files          84       84              
  Lines       18108    18115       +7     
==========================================
+ Hits        15761    15767       +6     
- Misses       2347     2348       +1     

Impacted Files	Coverage Δ
include/genn/genn/gennUtils.h	100.00% <100.00%> (ø)
src/genn/genn/synapseGroup.cc	85.51% <100.00%> (+0.05%)	⬆️
src/genn/backends/cuda/optimiser.cc	72.60% <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 09b74db...fa1c9f2. Read the comment docs.

[tnowotny]

Yes, makes sense. Nasty issues those ...