[QTI] Reusable legacy→QTI assessment-item conversion utility

[rtibbles]

Overview

Build the reusable legacy→QTI conversion — structured single/multiple selection, input, true/false → <qti-assessment-item> XML. This is the pure conversion utility; wiring it into the API endpoint (dual-read) is #6030, and the one-time global backfill is #6007.

Complexity: Medium
Target branch: unstable

Context

• Legacy items store structured question/answers with legacy types; the existing pydantic models already generate QTI for choice + text-entry — the legacy-expressible set.
• A uniform QTI model removes publish-time format branching and the editor's legacy-edit path ([QTI] Support legacy assessment item types in the QTI Editor #5980 becomes unnecessary).
• One conversion, three consumers: the API dual-read ([QTI] Serve legacy assessment items as QTI through the API endpoint (dual-read) #6030), ricecooker ingest ([QTI] Accept QTI items through the internal API for ricecooker import #6002), and the global backfill ([QTI] Backfill all non-raw AssessmentItems to QTI items #6007).

The Change

• Provide a reusable legacy→QTI conversion producing <qti-assessment-item> XML from a structured item (single/multiple selection, input, true_false).
• Make it lossless: prompt, choices/answers, correctness, hints, and media survive.
• Leave perseus_question items untouched — out of scope here.
• Surface items that fail conversion, rather than silently dropping them.

Acceptance Criteria

• A reusable conversion produces <qti-assessment-item> XML from each supported legacy type (single/multiple selection, input, true_false)
• The conversion is almost lossless: prompt, choices/answers, correctness, and media survive
• Each converted item validates against the QTI 3.0 XSD ([QTI] QTI 3.0 schema XML validation utilities #6005)
• Items that fail conversion are surfaced, not silently dropped
• Tests cover conversion per legacy type

Out of scope

• Hint conversion, this will be added in follow up

AI usage

Architecture decided with the maintainer across an iterative session: blanket QTI type with the item XML in raw_data; XSD-authoritative validation across all sources; a legacy→QTI global migration with an API-layer dual-read; ricecooker upload delegating to the AssessmentItem serializer; and a Perseus custom-interaction contract confirmed against the QTI 3.0 specification. Claude mapped the existing publish/validation/ricecooker code, proposed the breakdown, and drafted each issue; the maintainer steered every decision and reviewed throughout.

[rtibblesbot]

Watching for resolution on behalf of #6002.