Run canonical smoke tests with real benchmark sandboxes

[cm2435]

Summary

• switches canonical smoke fixtures from fake SmokePublicSandbox to each environment's real sandbox class
• renames root/dynamic smoke tasks to realistic semantic slugs and updates dashboard smoke expectations
• adds seeded source docs, resource handoff, coordination-thread checks, synthetic logprob telemetry, RL episode-view assertions, and one dynamic-subtask evaluator
• exercises production builtin toolkit surfaces from deterministic smoke workers
• adds a fast unit posture guard that rejects fake/test sandboxes in canonical smoke task snapshots
• marks live E2E smoke drivers xfail for real bugs exposed by the tighter posture rather than softening implementation logic

Known xfails

• private E2B templates may be unavailable in the executing account
• public object-bound sandbox calls do not currently emit command WAL like the old smoke-only InstrumentedSandbox path
• the current E2B detach path calls an SDK close() method not present on the locally installed AsyncSandbox object

Verification

• uv run ruff check tests/unit/smoke_base/test_smoke_real_sandbox_posture.py tests/e2e/_asserts.py tests/e2e/test_researchrubrics_smoke.py tests/e2e/test_minif2f_smoke.py tests/e2e/test_swebench_smoke.py tests/fixtures/smoke_components/smoke_base tests/fixtures/smoke_components/workers tests/fixtures/smoke_components/criteria tests/fixtures/smoke_components/benchmarks.py
• uv run ruff format --check tests/e2e/_asserts.py tests/fixtures/smoke_components/smoke_base/dynamic_tasks.py tests/fixtures/smoke_components/smoke_base/handoff_evaluator.py tests/unit/smoke_base/test_smoke_real_sandbox_posture.py
• uv run pytest tests/unit/smoke_base/test_smoke_real_sandbox_posture.py ergon_core/tests/unit/architecture/test_package_test_layout.py -q
• ERGON_DATABASE_URL=postgresql://ergon:ergon_dev@localhost:5433/ergon DATABASE_URL=postgresql://ergon:ergon_dev@localhost:5433/ergon INNGEST_API_BASE_URL=http://localhost:8289 ERGON_API_BASE_URL=http://127.0.0.1:9000 uv run pytest tests/e2e/test_researchrubrics_smoke.py::test_smoke_experiment_group tests/e2e/test_minif2f_smoke.py::test_smoke_experiment_group tests/e2e/test_swebench_smoke.py::test_smoke_experiment_group -q # 3 xfailed

[github-actions]

E2E smoke — researchrubrics

No PNG screenshots were uploaded for this leg. See screenshots/pr-133 for the uploaded placeholder.

[github-actions]

E2E smoke — swebench-verified

No PNG screenshots were uploaded for this leg. See screenshots/pr-133 for the uploaded placeholder.

[github-actions]

E2E smoke — minif2f

No PNG screenshots were uploaded for this leg. See screenshots/pr-133 for the uploaded placeholder.

[github-actions]

E2E smoke — swebench-verified

No PNG screenshots were uploaded for this leg. See screenshots/pr-133 for the uploaded placeholder.

[github-actions]

E2E smoke — minif2f

No PNG screenshots were uploaded for this leg. See screenshots/pr-133 for the uploaded placeholder.

[github-actions]

E2E smoke — researchrubrics

No PNG screenshots were uploaded for this leg. See screenshots/pr-133 for the uploaded placeholder.

[github-actions]

E2E smoke — researchrubrics

No PNG screenshots were uploaded for this leg. See screenshots/pr-133 for the uploaded placeholder.

[github-actions]

E2E smoke — swebench-verified

No PNG screenshots were uploaded for this leg. See screenshots/pr-133 for the uploaded placeholder.

[github-actions]

E2E smoke — minif2f

No PNG screenshots were uploaded for this leg. See screenshots/pr-133 for the uploaded placeholder.

[cm2435]

Closed as superseded by the new testing-posture cleanup stack: PR0 public sandbox cleanup plus PR1 stricter smoke/typed-WAL/RL assertions, to be rebased on the updated dev stack snapshot.