Background: In order to measure the value provided by Tricium, it would be good to measure the number of Tricium comments "acted upon". In order to measure that, we could compare the number of comments produced at different revisions (patchsets) in the change (CL).

For example, if we have the number of comments produced be [1, 1, 5, 3, 0], then we could assume that at some point Tricium found 5 issues and later they were all resolved. If it was [3, 3, 3] then we could infer that the author didn't consider those warnings useful or important enough to act on.

In order to do this, we would need to track changes and the list of revisions in each change.

Currently we do track the set of *currently open changes*, and the *latest revision* for each, but just for the purpose of polling (in tricium/appengine/gerrit/poll.go) - this could be expanded to track all changes (with an open/closed property) and all revisions for each change. A cron job could then later go through closed changes and compare numbers of comments across revisions.

Note, apart from collecting metrics, this would also enable a solution to bug 890545 -- before posting change, it would be possible to look up the run for the previous patchset and compare current comments against previous comments.