[Feature] Unit tests for intake extractor customer-id and section normalization

[sjohnston1972]

Part of the "Add an automated backend test suite" parent issue. Depends on the pytest-scaffolding sub-issue.

Background

backend/app/intake/extractor.py turns raw brief text into a structured {customer_id, customer_name, sections, overview, notes} dict. It has a pure regex path (extract_customer_id_regex) and an LLM path (extract_brief) that normalizes the model's JSON (fills all canonical sections, strips/caps the customer id, falls back to the regex).

Problem / Goal

Add unit tests for the extractor's deterministic behavior, stubbing the LLM router so tests stay offline.

Where to look

• backend/app/intake/extractor.py:35-46 — _CUSTOMER_RE, extract_customer_id_regex.
• extractor.py:66-128 — extract_brief: short-circuit for tiny input (< 20 chars), JSON-parse fallback, section normalization (CANONICAL_SECTIONS = ["LAN","WAN","EDGE","CLOUD","QOS"]), customer-id fallback to regex, 64-char cap.
• extractor.py:92-98 — it calls router.stream(...); monkeypatch this to yield canned output.

Suggested approach

Write backend/tests/test_extractor.py:

1. extract_customer_id_regex — matches Customer: ACME, customer - foo, is case-insensitive/multiline, strips trailing punctuation, returns None when absent.
2. extract_brief short-circuit: input under 20 chars returns the empty-sections stub with the "too short" note, and never calls the router.
3. extract_brief normalization: monkeypatch router.stream to yield a fenced JSON with only some sections; assert all five canonical keys are present and stripped.
4. Customer-id fallback: model returns customer_id: null but the text has a Customer: line → regex value is used; assert the 64-char cap.
5. Router returns unparseable text → the best-effort fallback (regex id, whole text in overview, failure note) is returned.

Use monkeypatch to replace router.stream with an async generator; mark tests async (scaffolding sub-issue enables pytest-asyncio).

Acceptance criteria

• test_extractor.py covers the five cases above and passes offline.
• No real LLM call is made (router is stubbed).

Testing

cd backend && pytest tests/test_extractor.py -q.

Out of scope

• The intake API endpoints and background tasks (api/intake.py) — logic-level extractor tests only here.