[engine-failure] PROJ-295-predicting-cognitive-load-from-eeg-spect @ tasks: RuntimeError

Automated report (spec 023 / FR-016) — an unexpected engine failure occurred while processing **PROJ-295-predicting-cognitive-load-from-eeg-spect** at stage `tasks`.

```
RuntimeError: LLMReviewer[constraint_preservation]: review frontmatter has neither `verdict:` nor `concerns:`/`action_items:` — a contentless/truncated review MUST NOT count as acceptance (false-convergence guard). First 200 chars: '```yaml\n---\nreviewer_name: constraint_preservation\nreviewer_kind: llm\nstage: tasks\nartifact_path: projects/PROJ-295-predicting-cognitive-load-from-eeg-spect/specs/001-predicting-cognitive-load-eeg/tas'
```

The project remains schedulable; it will retry on its normal cadence and recover once the underlying defect is fixed. Closing this issue does not require any project-state change.

Start at the LLMReviewer constraint_preservation handling for the tasks stage and inspect how review frontmatter is parsed. Reproduce the PROJ-295 failure or its normal retry, then verify that a contentless or truncated review cannot count as acceptance while the project remains schedulable.

Written by the indexing model from the issue text.