A red shard with two failing tests is no longer reported as a host crash

CI's test harness assumed xUnit v3 exits with the number of failing tests, so a shard whose host completed normally with two recorded failures (exit code 1) was classified as "host crashed after streaming results" and a synthetic HOST_CRASHED failure was written into the results — hiding the two real failures behind a crash that never happened. xUnit v3 exits 1 for any number of failures. The rule now reads: failures recorded and an exit code below the signal range means the host completed and the run is a plain test failure; a crash is a signal exit, a missing results file, or a non-zero exit with nothing recorded.