feat: upload wizard refactor + vision AI tags + artwork versioning

Upload wizard:
- Refactored UploadWizard into modular steps (Step1FileUpload, Step2Details, Step3Publish)
- Extracted reusable hooks: useUploadMachine, useFileValidation, useVisionTags
- Extracted reusable components: CategorySelector, ContentTypeSelector
- Added TagPicker component (studio-style list picker with AI badge + new-tag insertion)
- Fixed TagInput auto-open bug (hasFocusedRef guard)
- Replaced TagInput with TagPicker in UploadSidebar

Vision AI tag suggestions:
- Add UploadVisionSuggestController: sync POST /api/uploads/{id}/vision-suggest
- Calls vision.klevze.net/analyze/all on upload completion (before step 2 opens)
- Two-phase useVisionTags: immediate gateway call + background DB polling
- Trigger fires on uploadReady (not step change) so tags arrive before user sees step 2
- Added vision.gateway config block with VISION_GATEWAY_URL env

Artwork versioning system:
- ArtworkVersion / ArtworkVersionEvent models
- ArtworkVersioningService: createNewVersion, restoreVersion, rate limiting, ranking decay
- Migrations: artwork_versions, artwork_version_events, versioning columns on artworks
- Studio API routes: GET versions, POST restore/{version_id}
- Feature tests: ArtworkVersioningTest (13 cases)

config/vision.php

@@ -32,6 +32,18 @@ return [
     // Which derivative variant to send to vision services.
     'image_variant' => env('VISION_IMAGE_VARIANT', 'md'),
 
+    /*
+     |--------------------------------------------------------------------------
+     | Vision Gateway (aggregates CLIP + BLIP + YOLO via /analyze/all)
+     |--------------------------------------------------------------------------
+     | Falls back to CLIP base_url when VISION_GATEWAY_URL is not set.
+     */
+    'gateway' => [
+        'base_url'               => env('VISION_GATEWAY_URL', env('CLIP_BASE_URL', '')),
+        'timeout_seconds'        => (int) env('VISION_GATEWAY_TIMEOUT', 10),
+        'connect_timeout_seconds'=> (int) env('VISION_GATEWAY_CONNECT_TIMEOUT', 3),
+    ],
+
     /*
      |--------------------------------------------------------------------------
      | LM Studio – local multimodal inference (tag generation)