docs(website): sync content with the v3.4.0 conflict engine (#119)

The site described the engine as it was several releases ago. Audited every
engine-facing page against the real registry and fixed:

- Remove the phantom `section_only_change` pattern — it was listed as an
  auto-applied pattern on the landing page and conflict-engine page but has
  never existed in packages/core. Replaced with real patterns; the grids now
  show the true 12-entry registry with accurate auto/proposed/opt-in flags.
- Correct the hardcoded "10 patterns" count everywhere (landing 5 locales,
  conflict-engine, features, llms.txt, 5 compare pages, 1 blog post) to
  "8 deterministic patterns" (the auto-applied deterministic count) — the
  marketing number, distinct from the full registry.
- Normalize the 95% claim to "trivial conflicts" where it read "of merge
  conflicts" (index.md, conflict-engine.md); keep the figure.
- Document the v3.4.0 capabilities in guide/conflict-resolution.md:
  token_level_merge, refactoring_aware_merge, llm_proposed, the base-recovery
  pipeline step, and value_only_change's semver/timestamp-max resolution.
- core-api.md: fix "8 types", add the missing exports (summarizeTiers,
  resolveAsync, TokenMergeTrace, ResolutionTier, TierSummary, …).
- index.md JSON-LD softwareVersion 3.2.0 → 3.4.0, and add that field to
  bump-version.sh so it can't drift again.
- Refresh the hero announce + "New in vX" badge to v3.4 (5 locales).

Website build passes. (Left for a separate copy pass: the "What's New" tab
feature cards still show v2.x features.)

disabled

Co-authored-by: Laurent Guitton <laurent.guitton@dendreo.com>

Author: devlint

scripts/bump-version.sh

@@ -109,6 +109,13 @@ bump website/public/.well-known/mcp/server-card.json \
 bump packages/mcp/src/server.ts \
   "s/version: \"${OLD}\"/version: \"${NEW}\"/"
 
+# ── website/index.md (JSON-LD softwareVersion) ───────────────────────────────
+# Generic semver pattern: this field historically drifted (was pinned to an
+# older release than the rest) because it wasn't in the bump set — the generic
+# match fixes it regardless of what it currently holds.
+bump website/index.md \
+  "s/\"softwareVersion\": \"[0-9][0-9]*\.[0-9][0-9]*\.[0-9][0-9]*\"/\"softwareVersion\": \"${NEW}\"/"
+
 # ── README.md (badge URL) ────────────────────────────────────────────────────
 bump README.md \
   "s/version-${OLD}-/version-${NEW}-/"

website/.vitepress/theme/ConflictEnginePage.vue

@@ -1,14 +1,19 @@
 <script setup lang="ts">
+// 8 deterministic patterns auto-apply (auto: true). token_level_merge proposes
+// a merge you confirm; refactoring_aware_merge and llm_proposed are opt-in;
+// complex is the fallback that always hands the hunk back with its trace.
 const PATTERNS = [
   { name: 'same_change',           conf: 'certain', auto: true,  desc: 'Both branches made the exact same edit.' },
   { name: 'one_side_change',       conf: 'certain', auto: true,  desc: 'Only one branch touched this block.' },
+  { name: 'delete_no_change',      conf: 'certain', auto: true,  desc: 'One side deleted the block, the other left it untouched.' },
   { name: 'non_overlapping',       conf: 'high',    auto: true,  desc: 'Additions at different positions in the block.' },
   { name: 'whitespace_only',       conf: 'high',    auto: true,  desc: 'Same logic, different indentation or spacing.' },
   { name: 'reorder_only',          conf: 'high',    auto: true,  desc: 'Same lines, different order.' },
   { name: 'insertion_at_boundary', conf: 'high',    auto: true,  desc: 'New lines added at the edge of a hunk.' },
-  { name: 'value_only_change',     conf: 'high',    auto: true,  desc: 'A scalar value (JSON, config) updated on one side.' },
-  { name: 'section_only_change',   conf: 'high',    auto: true,  desc: 'A document section edited on one side only.' },
-  { name: 'llm_proposed',          conf: 'medium',  auto: true,  desc: 'LLM-proposed resolution above the confidence threshold.' },
+  { name: 'value_only_change',     conf: 'high',    auto: true,  desc: 'A scalar value (version, timestamp, hash) updated on both sides — keeps the higher semver / later timestamp.' },
+  { name: 'token_level_merge',     conf: 'medium',  auto: false, desc: 'Both sides changed disjoint tokens on the same line — proposes a merge you confirm, never auto-applied.' },
+  { name: 'refactoring_aware_merge', conf: 'high',  auto: false, desc: 'Rename/move detected and replayed across the conflict (opt-in).' },
+  { name: 'llm_proposed',          conf: 'medium',  auto: false, desc: 'AI-proposed resolution, validated post-merge (opt-in).' },
   { name: 'complex',               conf: 'low',     auto: false, desc: 'Overlapping edits — surfaced with full classification trace.' },
 ] as const
 </script>
@@ -19,7 +24,7 @@ const PATTERNS = [
       <div class="ph-inner">
         <span class="ph-badge">Conflict engine</span>
         <h1 class="ph-h1">Deterministic conflict resolution. <span class="grad">No guessing.</span></h1>
-        <p class="ph-sub">Every hunk runs through a classifier of 10 deterministic patterns, each with its own confidence profile and automatic resolver. The trivial 95% is resolved without you. The rest is surfaced with a full decision trace — never a black box.</p>
+        <p class="ph-sub">Every hunk runs through a classifier of pattern recognizers — 8 of them resolve deterministically on their own, each with its own confidence profile. The trivial 95% of conflicts is resolved without you. The rest is surfaced with a full decision trace — never a black box.</p>
         <div class="ph-ctas">
           <a href="/guide/conflict-resolution" class="ph-btn ph-btn--primary">Read the deep dive</a>
           <a href="/" class="ph-btn">← Back to home</a>
@@ -56,8 +61,8 @@ const PATTERNS = [
 
     <section class="ph-section ph-section--alt">
       <div class="ph-inner">
-        <h2 class="ph-h2">10 patterns. Deterministic. Auditable.</h2>
-        <p class="ph-secsub">Each pattern has its own confidence profile and automatic resolver. The classifier never guesses — when it can't be certain, it hands the hunk back to you with the trace.</p>
+        <h2 class="ph-h2">8 auto-applied patterns. Deterministic. Auditable.</h2>
+        <p class="ph-secsub">Eight deterministic patterns resolve on their own; the rest below either propose a merge you confirm, are opt-in, or hand the hunk back with its trace. The classifier never guesses — when it can't be certain, it stops.</p>
         <div class="pat-grid">
           <div v-for="p in PATTERNS" :key="p.name" class="pat" :class="{ 'pat--dim': !p.auto }">
             <div class="pat-head">

website/.vitepress/theme/FeaturesPage.vue

@@ -3,7 +3,7 @@ const GROUPS = [
   {
     title: 'Core Git',
     items: [
-      ['Smart resolution', '10 resolution patterns with a pattern registry and confidence scoring. 95%+ of trivial conflicts resolved without intervention.'],
+      ['Smart resolution', '8 deterministic patterns with a pattern registry and confidence scoring. 95%+ of trivial conflicts resolved without intervention.'],
       ['Integrated Pull Requests', 'Review GitHub, GitLab, Bitbucket and Azure DevOps PRs directly in the app — comments, reviews, CI status, conflict preview.'],
       ['Visual diff', 'Unified diff viewer with syntax highlighting, hunk-level staging, and merge preview.'],
       ['Folder tree diff', 'Flat ↔ tree toggle in the file list with per-folder aggregates, click-to-filter and a resizable sidebar.'],