feat(ai-service): full field set including line items

[Fluory]

Goal

Extraction returns every field the customer listed – header fields and line items – each with status and its own evidence, so a clerk can review a real quote request completely.

Acceptance criteria

• Header fields: company, contact person, e-mail, phone, requested delivery date, additional requirements
• Line items: description/product, quantity, unit, material, dimensions – each item field with its own status and evidence
• Output schema is versioned (schema_version); the prompt is a versioned file; both are stored with every extraction run
• Verifier normalises German number formats (decimal comma, thousands separators), units (mm, m, Stk., kg) and dates (DD.MM.YYYY, ISO); a calendar week ("KW 42") without an explicit date is at most uncertain
• A value whose quote cannot be found in the cited segment is unverified – also for line items
• TS side: contract types regenerated; persistence supports line items (item index) without breaking existing runs

Not part of this task

• New input formats (separate issue), review UI for line items (separate issue)

Affected areas

• services/ai/
• contracts/ai-service.openapi.yaml
• src/features/extraction/

Test plan

Criterion	Check
normalisers	pytest unit tests: numbers, units, dates, calendar weeks
verifier on line items	pytest unit tests
contract	contract test against OpenAPI
end to end	pytest with a recorded model response for a synthetic multi-item request

Security/Privacy affected?

Document content reaches the LLM – EU endpoint, synthetic data only, no content in logs.

Epic: #17 · Architecture: docs/decisions/ADR-0001-pilot-architecture.md

[Fluory]

Claimed by @Fluory (Claude Code, overnight run) on branch claude/feat-ai-full-fields-22 – stacked on #38 (claude/feat-user-management-30). The AI-service part ran in a parallel worktree while #30 finished; draft PR follows now.

---

Generated by Claude Code