fix: keep binary artifacts out of weekly brief text

[hamitarsln]

What and why

Weekly brief decoded every run artifact as UTF-8. A demo video stopped the report before the model call. Use the existing run metadata fallback for invalid UTF-8 or NUL bytes, and keep the artifact reference. Text still contributes its content; storage errors are not hidden.

The reporter also rejected a useful summary when the model omitted the main heading. Add a standard heading in that case. Keep an existing heading, remove model-written source lists, and reject an empty summary. Tin still appends the exact recorded sources.

Verification

• Earlier live runs failed in both projects: invalid byte 0x85 / 0xbd at position 35. Those runs used deployed code, not this branch.
• Local media fixtures cover both bytes and valid UTF-8 with NUL bytes. They now pass through source collection, report creation and artifact checks.
• Reproduced the missing-heading failure twice. Plain text, a subheading and a proper heading now work with one model call. A sources-only response is rejected.
• 47 local tests passed, no skips, with disposable PostgreSQL. Ruff and diff checks passed.

Limits

Provider and storage responses are synthetic. The fixed branch has not run against the deployed service; rerun both projects after rollout.

[Copilot]

🟢 Approval recommended

The focused implementation matches the documented behavior and includes adequate regression coverage.

0 open findings

What changed in this PR

Prevents binary run artifacts from breaking weekly brief generation while retaining source references and run metadata.

Changes:

• Falls back to run receipts for invalid UTF-8 or NUL-containing artifacts.
• Preserves text artifacts and propagates storage failures.
• Adds regression tests and documents the behavior.

File	Description
src/​tin_lite/​activities.py	Safely handles binary weekly brief sources.
tests/​test_weekly_brief_binary_sources.py	Covers binary, NUL, text, and storage-error cases.
docs/​engineering.md	Documents artifact fallback behavior.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.