Self-growing neural network, for possibly real AI applications

Connects any sensors and motors in a cluster, and generates a neuron network, to convert sensor inputs into motor actions.

Balance computing pressure among cluster, and deal with node drop & join.

The API of communicate with an existing neuron network, not necessarily be API between neurons.

1. (1b)flow: packet to be passed to any neurons connected to the receiving one
2. (1b)propagate：activate data storage, and effects connected neurons
3. (1b)ensure: whether to retry if failed
4. (5b)reserved

Packet type, including neuron management, data flow, etc. Neuron management includes query for neuron status, stop neuron, start neuron, delete neuron, create neuron, snapshot, etc.

Length of payload.

Address of target neuron, uuid seems to be overlong, we need an effective addressing scheme, that supports neuron drifting, and various underlying connection type.

Custom format according to type.

1. Unary operator: process data from one neuron, and pass on
2. Binary operator: add, subtract, mutiply, divide, ...
3. GUI app: client app to control neuron network, and for network monitoring
4. verification & encryption: stop hijacking in an open connecting environment
5. Matrix operator: even with deep learning nowadays
6. Matroid operator: final state, process on multi-modal input/output