Especially when we enable test selection at review time more broadly, this will save a bit of resources.

We just need to figure out how to detect a previous try push for the same patch from the autoland decision task.

Note: the fact that a given test was already run on try doesn't ensure that there would be no failure on autoland, since the patch might be rebased between the push to try and the landing. Failures due to bad rebases should be rare enough though.

Second note: we can also do the same when the same patch is pushed multiple times to try.

A potential solution that would require few changes in the taskgraph and support chunking in the taskgraph: implement this in the scheduler itself, the scheduler could get info from hg.mozilla.org and ignore the bugbug selected tests if they already run on try. So, the scheduler would 1) find the patch on hg.mozilla.org/try, 2) get all tests that run on that push; 3) from the bugbug selected tests, subtract the tests from 2.