Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@scripts/run_full_benchmark.py`:
- Around line 272-308: The timing.json currently sets "started_at" after all
targets run; use the earlier run_start to record the actual run start time:
convert run_start (captured before the loop) into an ISO 8601 UTC timestamp and
assign that to timing_data["started_at"] instead of datetime.now(...), keeping
total_seconds computed from total_elapsed; update the timing_data construction
(symbol: timing_data) to reference the start time derived from run_start and
ensure timezone is UTC and format matches .isoformat().
- Around line 156-168: Add explicit wall-clock start and finish timestamps
around the subprocess run so each target logs when it began and when it
completed/timeout; before setting t0 capture start_ts = datetime.now() and print
a "Started at {start_ts.isoformat()}" message (include cmd), and on both the
TimeoutExpired branch and the normal completion branch capture finish_ts =
datetime.now() and print "Finished at {finish_ts.isoformat()}" (alongside the
existing duration and cmd). Update the prints around variables cmd, t0, elapsed
and reuse _format_duration for duration output so both start and finish
timestamps (and durations) are logged for the function that runs the subprocess.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@scripts/run_full_benchmark.py`:
- Line 177: The main() function lacks a return type annotation; update its
signature to declare it returns an int (e.g., change def main(): to def main()
-> int:) and ensure all return paths within main (the places returning 0 or 1)
remain consistent with the int type annotation so type checkers and readers see
the intended return type.
- Around line 147-174: In _run, the timeout branch currently prints "Finished:"
which can be misleading; update the subprocess.TimeoutExpired except block to
print a clearer label (e.g., "Timed out at:" or "Timeout at:") instead of
"Finished:" so the console output matches the earlier "TIMEOUT" message — adjust
the print that uses finish_ts.strftime('%H:%M:%S UTC') in the exception handler
within _run to use the new label.

---

Duplicate comments:
In `@scripts/run_full_benchmark.py`:
- Around line 300-316: The timing.json lacks an explicit finished_at; compute
the end timestamp (either capture run_end_ts when the run completes or derive
run_end_ts = run_start_ts + timedelta(seconds=total_elapsed)) and add
"finished_at": run_end_ts.isoformat() into the timing_data dict alongside
"started_at" and "total_seconds" (update references to run_start_ts,
total_elapsed, and timing_data/timing_path accordingly so timing_path.write_text
writes the new field).