There are times when a failure to build a circuit through a relay is
not the relay's fault. See the comment added in this commit for more
rationalization.

Even after thinking about it for a few minutes and triple checking I was
thinking in the correct direction, I still managed to get it backwards.

A failure to measure a relay is not necessarily that relay's fault. Add
(and use) code that allows different ResultError subtypes to reduce a
result's freshness by different amounts. This way, if a bad result was
not likely to be the relay's fault, it won't be penalized very much.

GH: closes #61

GH: closes #106

ReadTheDocs doesn't install sbws so we can't import the version.
So we have to essentially grep for it.

GH: closes #100

GH: closes #97

If it happens in the config validation, we cannot init sbws since we'll
fail to validate.

Somewhat reverts c52a850b

GH: closes #95

GH: closes #90