feat: improve AI resilience, add background enrichment loading states, and fix duplicate identifier in TagSelector

.aiideas

@@ -1,28 +1,85 @@
-Act as a Senior TypeScript/Next.js Developer.
+´
 
-I need a robust client-side image processing utility (an "Image Agent") to optimize user uploads before sending them to an LLM or Supabase.
+Act as a Senior Next.js Developer.
 
-**Task:**
-Create a utility file `src/utils/image-processing.ts`.
-This file should export a function `processImageForAI` that uses the library `browser-image-compression`.
+Refactor the whisky analysis logic to optimize for perceived performance using a "Optimistic UI" approach.
+Split the current `analyzeBottle` logic into **two separate Server Actions**.
 
-**Requirements:**
-1.  **Input:** The function takes a raw `File` object (from an HTML input).
-2.  **Processing Logic:**
-    - Resize the image to a maximum of **1024x1024** pixels (maintain aspect ratio).
-    - Convert the image to **WebP** format.
-    - Limit the file size to approx **0.4MB**.
-    - Enable `useWebWorker: true` to prevent UI freezing.
-3.  **Output:** The function must return a Promise that resolves to an object with this interface:
-    ```typescript
-    interface ProcessedImage {
-      file: File;           // The compressed WebP file (ready for Supabase storage)
-      base64: string;       // The Base64 string (ready for LLM API calls)
-      originalFile: File;   // Pass through the original file
-    }
-    ```
-4.  **Helper:** Include a helper function to convert the resulting Blob/File to a Base64 string correctly.
-5.  **Edge Cases:** Handle errors gracefully (try/catch) and ensure the returned `file` has the correct `.webp` extension and mime type.
+### 1. Create `src/app/actions/scan-label.ts` (Fast OCR)
+This action handles the image upload via `FormData`.
+It must use the model with `safetySettings: BLOCK_NONE`.
+It must **NOT** generate flavor tags or search strings.
 
-**Step 1:** Give me the `npm install` command to add the necessary library.
-**Step 2:** Write the complete `src/utils/image-processing.ts` code with proper JSDoc comments.
+**System Prompt for this Action:**
+```text
+ROLE: High-Precision OCR Engine for Whisky Labels.
+OBJECTIVE: Extract visible metadata strictly from the image.
+SPEED PRIORITY: Do NOT analyze flavor. Do NOT provide descriptions. Do NOT add tags.
+
+TASK:
+1. Identify if the image contains a whisky/spirit bottle.
+2. Extract the following technical details into the JSON schema below.
+3. If a value is not visible or cannot be inferred with high certainty, use null.
+
+EXTRACTION RULES:
+- Name: Combine Distillery + Age + Edition + Vintage (e.g., "Signatory Vintage Ben Nevis 2019 4 Year Old").
+- Distillery: The producer of the spirit.
+- Bottler: Independent bottler (e.g., "Signatory", "Gordon & MacPhail") if applicable.
+- Batch Info: Capture ALL Cask numbers, Batch IDs, Bottle numbers, Cask Types (e.g., "Refill Oloroso Sherry Butt, Bottle 1135").
+- Codes: Look for laser codes etched on glass/label (e.g., "L20394...").
+- Dates: Distinguish clearly between Vintage (distilled year), Bottled year, and Age.
+
+OUTPUT SCHEMA (Strict JSON):
+{
+  "name": "string",
+  "distillery": "string",
+  "bottler": "stringOrNull",
+  "category": "string (e.g. Single Malt Scotch Whisky)",
+  "abv": numberOrNull,
+  "age": numberOrNull,
+  "vintage": "stringOrNull",
+  "distilled_at": "stringOrNull (Year/Date)",
+  "bottled_at": "stringOrNull (Year/Date)",
+  "batch_info": "stringOrNull",
+  "bottleCode": "stringOrNull",
+  "whiskybaseId": "stringOrNull",
+  "is_whisky": boolean,
+  "confidence": number
+}
+
+2. Create src/app/actions/enrich-data.ts (Magic/Tags)
+
+This action takes name and distillery (strings) as input. No image upload. It uses gemini-1.5-flash to retrieve knowledge-based data.
+
+System Prompt for this Action:
+Plaintext
+
+TASK: You are a Whisky Sommelier.
+INPUT: A whisky named "${name}" from distillery "${distillery}".
+
+1. DATABASE LOOKUP:
+Retrieve the sensory profile and specific Whiskybase search string for this bottling.
+
+2. TAGGING:
+Select the top 5-8 flavor tags strictly from this list:
+[${availableTags}]
+
+3. SEARCH STRING:
+Create a precise search string for Whiskybase using: "site:whiskybase.com [Distillery] [Vintage/Age] [Bottler/Edition]"
+
+OUTPUT JSON:
+{
+  "suggested_tags": ["tag1", "tag2", "tag3"],
+  "suggested_custom_tags": ["unique_note_if_missing_in_list"],
+  "search_string": "string"
+}
+
+3. Integration
+
+Update the frontend component to:
+
+    Call scan-label first.
+
+    Update the UI state with the metadata immediately.
+
+    If the scan was successful, automatically trigger enrich-data in the background to fetch tags and search string.