PageRank: Pass-by nodes

[teamdandelion]

I think we'll need to add a generalization to PageRank, which is a way for cred to "pass-thru" a node.

Consider a graph with nodes C1...C10 (10 regular contributions), W (a wrapper), D (an important dependency). The edges are as follows:

C{1..10} -> W
W -> D

Then we can expect the cred ordering to be `[W, D, C1...C10].

However, now imagine that C1...C10 represent contributions in a project, D represents a very complex/valuable dependency that they all depend on, and W is a trivial wrapper over D. Then giving W more cred than D is clearly wrong. However, the naive solution--to reduce the weight on W--also prevents cred from appropriately flowing on to D.

Instead, we need a way for the edges traveling from C1...C10 to transfer an undiminished amount of cred to W, but for most of that cred to "pass by" W and then arrive at D. It's sort of like each node has a "stickiness" for keeping cred, which is independent of its "gravity" for attracting cred. This is in contrast to our current approach for node weights, wherein a node weight modifies the amount of cred that flows over every incident connection, so having a high node weight is like having a high "gravity" in that nearby cred tends to flow towards the highly weighted node.

This could also be useful for solving the identity resolution problem: perhaps the various identities people acquire (GitHub handle, email address, etc) can be "pass-by" nodes that are connected to the "core" identity which is a deduplicated representation of any particular person.

Thanks to @wchargin for this idea.