docs(api): document hash-conflict resolution endpoints

Cover the admin API added in 03cb4c7 for duplicate-content decisions:

- GET /api/admin/hash-conflicts — pending conflict groups with member
  items and per-item usage counts (progress, highlights, bookmarks,
  notes, collections)
- POST /api/admin/hash-conflicts/:id/resolve — action=keep (merge child
  rows into keep_uuid, delete losers) vs action=keep_all (dismiss);
  JSON and form-encoded bodies, error codes including 409 for already
  resolved
- When conflicts are created (startup backfill, rescans) and the
  guarantee that files on disk are never deleted

docs/developer/api/admin/hash-conflicts.md

@@ -0,0 +1,111 @@
+# Hash Conflicts API
+
+## Overview
+
+When Bookhoard hashes your library (on import, rescan, or the startup backfill), two media items in the same library with the same `file_sha256` indicate duplicate content. Each duplicate group is recorded as a **hash conflict** and exposed here for an explicit keep/merge decision. Conflicts are also surfaced in the admin UI's Hash Conflicts page.
+
+**Authentication**: Admin JWT token required
+**Content-Type**: `application/json` (resolve also accepts form-encoded bodies for htmx)
+
+---
+
+## Endpoints
+
+### List Hash Conflicts
+
+List all pending conflict groups, each with its member items and per-item usage counts (reading progress, highlights, bookmarks, notes, collections) to help decide which copy to keep.
+
+**Endpoint**: `GET /api/admin/hash-conflicts`
+
+**Response**: **200 OK**
+
+```json
+{
+  "conflicts": [
+    {
+      "id": "conflict-uuid",
+      "library_id": "library-uuid",
+      "library_name": "Ebooks",
+      "sha256": "abc123...",
+      "created_at": "2026-08-14T12:00:00Z",
+      "items": [
+        {
+          "id": "media-item-uuid",
+          "title": "The Hobbit",
+          "author": "J. R. R. Tolkien",
+          "file_path": "/books/hobbit.epub",
+          "file_size": 1048576,
+          "created_at": "2026-01-01T00:00:00Z",
+          "progress_count": 2,
+          "highlight_count": 12,
+          "bookmark_count": 3,
+          "note_count": 1,
+          "collection_count": 2
+        }
+      ]
+    }
+  ],
+  "total": 1
+}
+```
+
+**Example**:
+
+```bash
+curl -X GET https://bookhoard.example.com/api/admin/hash-conflicts \
+  -H "Authorization: Bearer <admin_token>"
+```
+
+---
+
+### Resolve Hash Conflict
+
+Resolve one conflict group.
+
+**Endpoint**: `POST /api/admin/hash-conflicts/{id}/resolve`
+
+**Request Body** (JSON or form-encoded):
+
+```json
+{
+  "action": "keep",
+  "keep_uuid": "media-item-uuid-to-keep"
+}
+```
+
+| Field       | Type   | Required             | Description                                                                                                                     |
+| ----------- | ------ | -------------------- | ------------------------------------------------------------------------------------------------------------------------------- |
+| `action`    | string | Yes                  | `keep_all` — both copies are intentional; dismiss the conflict. `keep` — keep `keep_uuid` and delete the other copies.          |
+| `keep_uuid` | string | for `action=keep`    | The media item UUID to keep. Must belong to this conflict group. With `keep`, every other copy's child rows (progress, highlights, bookmarks, notes, collections, …) are merged into the kept item before the losers are deleted. |
+
+**Responses**:
+
+- `200 OK` — resolved (body is an HTML confirmation snippet for the admin UI page)
+- `400 Bad Request` — invalid conflict ID, missing `keep_uuid`, or `keep_uuid` not in the group
+- `404 Not Found` — conflict doesn't exist
+- `409 Conflict` — conflict already resolved
+
+**Example**:
+
+```bash
+curl -X POST https://bookhoard.example.com/api/admin/hash-conflicts/<id>/resolve \
+  -H "Authorization: Bearer <admin_token>" \
+  -H "Content-Type: application/json" \
+  -d '{"action": "keep", "keep_uuid": "media-item-uuid"}'
+```
+
+---
+
+## When Conflicts Are Created
+
+- **Startup backfill**: items imported before hashing existed are hashed automatically ~30s after startup; duplicates discovered land here.
+- **Rescan**: hashes are recomputed and content duplicates are flagged.
+
+Files on disk are never deleted — resolution only affects database rows.
+
+---
+
+## Related Endpoints
+
+- [System Settings API](../system/settings.md) — scanning configuration
+- [Scanner API](../scanner/) — triggering scans and watch mode