feat: phase 2 of AI pipeline hardening — tool-based structured outputs + prompt caching

Every structured-output call now uses an Anthropic tool instead of
parsing JSON out of free-form prose, and stable system prompts are
sent as cacheable blocks. Behaviour-equivalent to phase 1 from the
caller's point of view; the savings show up in token usage and in the
absence of "AI returned non-JSON response" failure modes.

* src/lib/llmTools.js — single source of truth for tool definitions:
  emit_knowledge_graph, emit_handbook_delta, emit_learning_article /
  _slides / _infographic / _all, emit_custom_topic, emit_quiz_questions,
  emit_graph_actions, plus five article-patch tools (set_intro,
  set_section, add_section, remove_section, replace_takeaways).
* src/lib/articlePatches.js — pure applyArticlePatches +
  applyAndValidate; rebuilds the article from a sequence of patch tool
  calls and re-validates against learningArticleSchema. set_section
  falls back to appending when no matching heading exists so the
  model's intent is preserved rather than silently dropped.
* src/lib/llmSchemas.js — Zod schemas for the five patch ops,
  registered in toolSchemaRegistry so callLLM validates them
  automatically.
* src/lib/llm.js — simulation mode now returns a tool_use stub matching
  toolChoice.name, so the UI keeps working with Simulation Mode on
  after the structured-output migration.
* src/lib/extractionPipeline.js — processSourceText and
  analyzeHandbookDelta migrated to callLLM + tool use. System prompts
  sent as { cache_control: ephemeral } blocks. Handbook results pass
  through normalizeHandbookResult to collapse legacy "executes"
  relations into executed_by with swapped source/target.
* src/lib/learningService.js — generateLearningContent picks the right
  tool per selectedType; generateCustomTopic uses emit_custom_topic;
  refineLearningContent now drives the five patch tools with
  toolChoice 'any' and rejects the whole turn if the patched article
  fails validation. Article-only refinement is intentional for phase 2;
  refining a topic without an article surfaces a clear error.
* src/lib/testService.js — quiz generation via emit_quiz_questions.
* src/components/admin/KnowledgeGraph.jsx — analyzeGraph routed through
  the reasoning tier (Opus) since graph-wide consolidation benefits
  from a stronger reasoner.
* src/components/chat/prompts.js — buildSystemPrompt now returns three
  text blocks: stable preamble (cached), KB context (cached, hash-bust
  deferred to phase 5), per-turn user/admin tail (uncached).
* src/lib/__tests__/ — 13 new tests covering each patch op, multi-op
  sequencing, post-patch validation failure, and tool/registry shape.

Acceptance: lint and 45/45 tests green; build succeeds; no
`match(/\{[\s\S]*\}/)` JSON extraction left in src/. Live verification
of cache hits on a second extraction within 5 minutes is deferred to
manual smoke testing — needs real `/api/anthropic` traffic.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/lib/llm.js

@@ -125,7 +125,7 @@ function isChatLikeTask(task) {
   return task === 'legacy.chat' || task.startsWith('chat.') || task.startsWith('r42.');
 }
 
-const SIMULATION_EXTRACTION_PAYLOAD = JSON.stringify({
+const SIMULATION_EXTRACTION_GRAPH = {
   topics: [
     { id: 'radicale-transparantie', label: 'Radicale Transparantie', type: 'concept', description: 'De kernwaarde van Respellion waarbij alle informatie publiek toegankelijk is.', learning_relevance: 'core' },
     { id: 'kennisbeheer', label: 'Kennisbeheer', type: 'process', description: 'Het proces van het vastleggen en ontsluiten van organisatiekennis.', learning_relevance: 'standard' },
@@ -135,28 +135,66 @@ const SIMULATION_EXTRACTION_PAYLOAD = JSON.stringify({
     { source: 'kennisbeheer', target: 'radicale-transparantie', type: 'depends_on' },
     { source: 'wekelijkse-sessie', target: 'kennisbeheer', type: 'part_of' },
   ],
-});
+};
+
+const SIMULATION_EXTRACTION_PAYLOAD = JSON.stringify(SIMULATION_EXTRACTION_GRAPH);
 
 const SIMULATION_CHAT_TEXT =
   'Simulatiemodus staat aan — vraag een beheerder om Simulation Mode uit te zetten in Admin → Settings om met R42 te chatten.';
 
-async function simulatedResponse({ task }) {
-  await new Promise((r) => setTimeout(r, 400));
-  if (isChatLikeTask(task)) {
-    return {
-      text: SIMULATION_CHAT_TEXT,
-      toolUses: [],
-      stopReason: 'end_turn',
-      usage: { input_tokens: 0, output_tokens: 0, cache_creation_input_tokens: 0, cache_read_input_tokens: 0 },
-      requestId: null,
-      model: 'simulation',
-      durationMs: 400,
-    };
-  }
+const SIMULATION_ARTICLE = {
+  title: 'Voorbeeld leermodule',
+  intro: 'Dit is een simulatie. Schakel Simulation Mode uit om echte content te genereren.',
+  sections: [
+    { heading: 'Wat dit is', body: 'Dit is een placeholder-sectie die alleen verschijnt wanneer simulatiemodus aan staat. Hij illustreert de structuur van het artikel zonder een echte API-aanroep te doen. Dat is handig voor UI-werk.' },
+  ],
+  keyTakeaways: ['Simulatiemodus levert geen echte inhoud.', 'Schakel uit voor productie.'],
+};
+
+const SIMULATION_SLIDE = {
+  title: 'Voorbeeldslide',
+  bullets: ['Eerste punt', 'Tweede punt'],
+  speakerNote: 'Spreker-notitie ter illustratie.',
+};
+
+const SIMULATION_INFOGRAPHIC = {
+  headline: 'Simulatie',
+  tagline: 'Vervang door echte content',
+  stats: [{ value: '100%', label: 'simulatie', icon: '📊' }],
+  steps: [{ number: 1, title: 'Schakel uit', description: 'Zet simulatiemodus uit in Admin → Settings.', icon: '🔧' }],
+  quote: 'Een simulatie vertelt niets nieuws.',
+  colorTheme: 'teal',
+};
+
+const SIMULATION_TOOL_STUBS = {
+  emit_knowledge_graph: SIMULATION_EXTRACTION_GRAPH,
+  emit_handbook_delta: SIMULATION_EXTRACTION_GRAPH,
+  emit_learning_article: { article: SIMULATION_ARTICLE },
+  emit_learning_slides: { slides: [SIMULATION_SLIDE] },
+  emit_learning_infographic: { infographic: SIMULATION_INFOGRAPHIC },
+  emit_learning_all: { article: SIMULATION_ARTICLE, slides: [SIMULATION_SLIDE], infographic: SIMULATION_INFOGRAPHIC },
+  emit_custom_topic: { label: 'Simulatie onderwerp', type: 'concept', description: 'Een placeholder-onderwerp gegenereerd in simulatiemodus.' },
+  emit_quiz_questions: {
+    questions: [
+      {
+        id: 'sim-q1',
+        question: 'Wat doet simulatiemodus?',
+        topicLabel: 'Simulatie',
+        options: ['Echte API-aanroepen', 'Stub-data tonen', 'Niets', 'Crasht de app'],
+        correctIndex: 1,
+        explanation: 'Simulatiemodus retourneert vaste stub-data zonder de API te raken.',
+      },
+    ],
+  },
+  emit_graph_actions: { merges: [], deletions: [], newRelations: [], relevanceUpdates: [] },
+  set_intro: { intro: 'Bijgewerkte intro (simulatie).' },
+};
+
+function stubResponse({ stopReason = 'end_turn', text = '', toolUses = [] }) {
   return {
-    text: SIMULATION_EXTRACTION_PAYLOAD,
-    toolUses: [],
-    stopReason: 'end_turn',
+    text,
+    toolUses,
+    stopReason,
     usage: { input_tokens: 0, output_tokens: 0, cache_creation_input_tokens: 0, cache_read_input_tokens: 0 },
     requestId: null,
     model: 'simulation',
@@ -164,6 +202,22 @@ async function simulatedResponse({ task }) {
   };
 }
 
+async function simulatedResponse({ task, toolChoice }) {
+  await new Promise((r) => setTimeout(r, 400));
+
+  if (toolChoice?.type === 'tool' && SIMULATION_TOOL_STUBS[toolChoice.name]) {
+    return stubResponse({
+      stopReason: 'tool_use',
+      toolUses: [{ name: toolChoice.name, input: SIMULATION_TOOL_STUBS[toolChoice.name] }],
+    });
+  }
+
+  if (isChatLikeTask(task)) {
+    return stubResponse({ text: SIMULATION_CHAT_TEXT });
+  }
+  return stubResponse({ text: SIMULATION_EXTRACTION_PAYLOAD });
+}
+
 function linkSignals(userSignal, timeoutSignal) {
   const controller = new AbortController();
   const abort = (reason) => controller.abort(reason);
@@ -241,7 +295,7 @@ export async function callLLM(options) {
   if (!task) throw new Error('callLLM requires a `task` label.');
 
   const useSimulation = storage.get('admin:use_simulation') === true;
-  if (useSimulation) return simulatedResponse({ task });
+  if (useSimulation) return simulatedResponse({ task, toolChoice });
 
   const model = resolveModel(tier);
   const messagesPayload = buildMessages({ messages, user });