Client can decide which url to connect to

[asmodehn]

Currently we use the 'name' as a unique id and discovery primitive.
And we rely on an external mechanism transmitting the name to another process, to be able to do discovery.

This is bad for few reasons :

• naming a node tend to make people give "responsibilities" to one node / process, making the structure quite brittle (one process crash, then nothing relying on this feature works aymore...)
• it introduce an intermediary, where we could directly use the ZMQ url (or any network uid primitive depending on our network implementation.
• we do not have a clear API for the user to make it obvious how to discover a running system.

Before we can do complex things like auto discovery, we need to build a simple and obvious API for "manual discovery"

[asmodehn]

So we should probably :

• drop the "name" concept for a node, assigning uuids only.
• when starting a node, it should return the socket address url that a client can use to later discover the services available on that node.
• nodes url should be propagated up the startign tree : a node starting another node dynamically somehow needs to propagate the url to connect to the nodes he started.

As a result a node is supposed to manage his own children for start / stop / exception / error / etc.
Although the structure is also a tree, this does NOT match the loggers structures, as the loggers should be defined by functionalities (ie : my.special.algorithm) where the nodes should be entirely redundant and replaceable (ie: 123252.242342.33438)

=> This implies that our logging system (TODO) needs to weave multiple nodes output in the same log file, while dealing with concurrency issues.