Restore timeout-warning trace parity with legacy in concrete mode

What does this MR do?

Brings concrete-mode runner trace output to parity with legacy for timeout and after-script-failure scenarios. Four commits:

1. Honor AFTER_SCRIPT_IGNORE_ERRORS in concrete mode — the concrete builder forced AllowFailure: true on the after-script step, causing functions/concrete/run/stages/step.go:41-44 to swallow every after-script error before the runner-level AFTER_SCRIPT_IGNORE_ERRORS gate at functions/concrete/run/runner.go:271-273 could see it. The user-facing variable was a silent no-op in concrete mode. Removes the redundant override; matches legacy common/build.go:905-919. Also updates TestBuildConcreteKitchenSink's expected JSON to drop the allow_failure field from the after-script step (no longer set, so json:omitempty omits it).
2. Emit step_script timeout warning in concrete mode — restores the legacy warning emitted by common/build.go:632-639 when the script context trips a deadline.
3. Emit after_script timeout warning in concrete mode — restores the matching legacy warning for after-script.
4. Emit job-level timeout warning in concrete dispatch path — concrete dispatches the entire job as one monolithic "concrete" build stage at common/build.go:782, bypassing both executeStage's per-stage defer (line 632-639) and the legacy step-dispatch post-call check (line 612-616). Concrete's own per-phase warnings (commits 2 and 3) reach the trace for inner-context timeouts but are lost on a job-level deadline because the build orchestrator force-kills the step-runner container before stderr drains. Emits a phase-agnostic `Job could not run to completion …` from the build orchestrator after executeStepStage returns, gated on errors.Is(ctx.Err(), context.DeadlineExceeded) — covers the job-level timeout case that commits 2 and 3 cannot reach.

Why was this MR needed?

The control flow in concrete mode (failure_reason, CI_JOB_STATUS, exit codes) was already correct on timeout — only the user-visible trace warnings that legacy emits were missing, and as a side effect of the redundant AllowFailure: true override on after-script, `AFTER_SCRIPT_IGNORE_ERRORS=false` was also silently ignored in concrete mode, diverging from the documented user-facing contract.

What's the best way to test this MR?

• Locally: go test ./functions/concrete/... and go test ./common/...
• End-to-end (pilot-runners): build a runner image from this branch and point pilot-runners at it. After commits 1–3 the phase-level timeout tests (timeout/script_timeout, timeout/after_script_timeout_but_job_is_successful) pass under `runner-concrete-test`. After commit 4 the two job-level timeout cases (timeout/job_timeout_during_script, timeout/job_timeout_during_after_script_and_job_is_unsuccessful) produce the new phase-agnostic warning in `job.trace`; the matching runner-backbench assertion update for those two is tracked separately.

What are the relevant issue numbers?

N/A