Properly control validity of NodeCollections and SynapseCollections with hashes

[heplesser]

NodeCollections and SynapseCollections remain existent in Python even after a ResetKernel(). We currently protect against failing C++ assertions on set()/get() on SynapseCollection by checking against an empty network, but that is approximate at best.

A robust solution would be to generate a time-based hash every time ResetKernel() is called. Each NodeCollection is assigned the kernel hash valid when it is created, and any operation using the NodeCollection to interact with the kernel checks that the NC-hash is equal to the current kernel hash.

For SynapseCollections, one can do things correspondingly, but here a new hash should be set every time connections_have_changed is set to true. The "no longer valid" warning can then disappear, as the user will get an error.

[github-actions]

Issue automatically marked stale!

[clinssen]

This would solve an issue with slowdowns happening due to repeatedly setting a parameter of a SynapseCollection.

@heplesser tracked this down to the following.

Specifically, in pynest/nest/lib/hl_api_types.py, SynapseCollections.get() and .set() both contain

   if self.__len__() == 0 or GetKernelStatus("network_size") == 0:

The underlying problem is that GetKernelStatus() first collects the full network status and then picks out the network size. Again, I have no idea why this took about 75% of the total runtime and why it got slower over time (there is not more data to return).