Deadline auto-holds in submit_decisions() are still silent

[FlyM1ss]

Follow-up from #374 (T2 of the burst-capacity-safety plan), which made deadline auto-holds
visible at the three sites that go through _maybe_apply_timeout.

ExternalBacktestSession.submit_decisions() has a fourth, independent deadline check that
bypasses _maybe_apply_timeout entirely, so the hold it creates prints nothing:

# dashboard/backend/domain/backtesting/external_run_service.py:~514
if _utcnow() > self._deadline_at():
    self._advance_step(executable=[], decision_source="timeout_hold")

Traced all three production callers of session.submit_decisions(...):

caller	preceded by an instrumented call?
domain/runs/service.py:909 — /api/v1/runs/* (shipping SDK)	yes — get_status() pre-empts this branch under the same lock
api/routers/external_backtest.py:212 — legacy /api/v1/backtest/* (Discord bot, built-in agents)	no — silent
api/v2/runs.py:303 → execution/backtest_backend.py:226 — canonical v2 contract	no — silent

Both silent paths are ordinary GET-then-POST flows where the deadline elapses in the gap, which
is exactly the case the feature exists to surface. The periodic reaper does not cover it either:
submit_decisions has already advanced the step, so drain_expired finds nothing expired.

Suggested fix: call the same emit helper #374 added, from this branch too — a couple of lines,
no wire-contract change. Left out of #374 deliberately because the plan scoped that task to
exactly three sites.