Sub-optimal merging of synapse groups in models with homogeneous delays

[neworderofjamie]

SynapseGroup::getWUHashDigest() is used as the basis for merging all types of synapse group but this includes the hash of both getSrcNeuronGroup()->getNumDelaySlots() and getTrgNeuronGroup()->getNumDelaySlots(). The latter is not necessary in synapse dynamics or presynaptic updates unless a postsynaptic variable is accessed but means that in models with N homogeneous delays, you end up with N^2 merged synapse groups.

[neworderofjamie]

Also, all weight update model parameters are compared even if they aren't referenced e.g. in a spike-like event threshold