Change the way neuron values are cached in NeuralNetwork.

[HyperCodec]

By moving the cache to a separate HashMap or something, neurons would not need to be stored in a RwLock, greatly reducing the amount of data being cloned while transferring between topology and mutable. Additionally, flush_state could be merged into predict without fear of performance issues because the map can simply be replaced by a new one in memory, which is way faster than the current iterative approach.

[HyperCodec]

Probably also helps #63 because the neuron types could become one.

[HyperCodec]

The main difficulty I see with this approach is accidentally wasting CPU cycles due to situations where a neuron is in the process of being solved but other threads also try to solve it because there is no RwLock telling it that the work is already being done. Perhaps this could be solved by first creating a temporary locking entry that forces the other ones to wait until it has finished before having free read access to the value.