What was the problem?

In our larger effort to remove python indirection from cython runtime , there were some changes left for the synapses_create_array and synapses_create_generator templates.

Every time we created new synapses, the fast Cython code had to stop and call back to Python (_owner._resize() and _owner._update_synapse_numbers()). This Python step was in charge of resizing all the synaptic arrays (like weights, delays, etc.) and updating counters like N_incoming. This back-and-forth between Cython and Python is super slow and was a real bottleneck.

What's the fix?

I've moved all that logic directly into the Cython templates, just like how standalone mode already does it.

This means when you create synapses now:

• All the registered arrays get resized immediately in C++ (varname_ptr.resize(...)).
• The N_incoming, N_outgoing, and total N counters are updated right there in Cython.
• The synapse_number (for multisynaptic connections) is also calculated in Cython.

Note : This is still WIP as I need to verify a few cases and also do the changes ofr the synapses_create_generator template