account for tests from a missing shard by marking them as notrun
When a shard machine does not start IGT execution the whole shard is lost, i.e.: the tests that were in the list were not accounted for, even as "notruns".
That causes the final test count to wildly vary in increments of ~80 which makes things harder to track.
One way of assuring that we have stable test counts (+/- the normal evolution of the tests) is to generate the list of missing tests by running
igt_runner --dryrun for each missing "shard chunk" during the result collection phase.