fix(blend): ensure all nodes are connected and have the same peering degree

[youngjoon-lee]

For accurate simulation, we need to ensure that:

• All nodes are connected (There is no network partition).
• All nodes have the same number of connections, as specified as peering degree.

Of course, the real world won't be as ideal as this simulation. But, as the first step, we need to make the uniform/predictable configuration to easily interpret the result of simulation. Later, we can take it further.

[danielSanchezQ]

It looks good.
But what is the target of topologies?
Are we gonna explore more on how to construct them?
Is this the final chosen algorithm?

[youngjoon-lee]

It looks good. But what is the target of topologies? Are we gonna explore more on how to construct them? Is this the final chosen algorithm?

In production, we're going to let each node choose nodes that they want. We are not going to define any algorithm to build a topology.
However, here, by using the simulation, first of all, I'd like to see the impact of each parameter (network size, peering degree, etc.) on latency and bandwidth. If the topology is built 100% randomly (as implemented in the master branch), there are sometime one or two nodes who have more connections than others. Or, rarely, the network is partitioned. Then, we'll get some peak numbers in the latency results, or non-uniform bandwidth results, which makes us hard to understand the impact of parameters. Therefore, I'd like to make a topology as uniform as possible for now. Later, if necessary, we can run simulations with various shape of topologies, but not now.

Plus, in the specification, we defined that all connections are bidirectional. But the simulation in the master branch was different from the spec. Each node chose random peers for only outbound connections, and sent messages, but the peers don't send messages to the node. They chose peers independently for outbound connections. It's a bug, and this PR fixes it.