fix(harness): preserve per-service test accounting on timeout

[innocentmchry]

When a harness container timed out, timeout handling could propagate a non-usable status into aggregation.
That could short-circuit to a single fallback failure for the whole datapoint, dropping other service-level
results from accounting. In multi-service cases this undercounted total tests (often removing passes) and
skewed the final pass rate.

• Return an explicit non-zero timeout status (124) for timed-out runs.
• Count service outcomes with binary semantics (exit_code != 0) instead of summing raw exit codes.

This makes reporting consistent: each service contributes one test result, timed-out services are failed,
and total/passed/failed numbers reflect all executed service outcomes.