[tor-relays] write-history for exit relays only?

[Aaron Johnson]

> I suspect that one could approximate this number by accounting for the
> probability of all exits being selected as guard, middle, and exit, but
> I would prefer a simpler and more reliable approach.

This doesn’t seem like a bad approximation to me, given that for as long as I have been aware, exits have had zero probability of being chosen in any position other than the exit position.

Aaron