feat: recherche amelioree - scoring fuzzy multi-niveaux (#7)

- Algorithme de scoring: exact > starts-with > substring > all words > any word
- Scores: 1.0 > 0.95 > 0.85 > 0.7 > 0.5 > 0.3
- Tolérance aux fautes de frappe via matching partiel sur mots
- Résultats triés par pertinence décroissante
- Supporte les titres en français, anglais, romaji

Closes #7

app/routers/router_anime.py

@@ -174,10 +174,28 @@ async def search_anime_unified(
 
             if url and url not in seen_urls:
                 seen_urls.add(url)
-                if q.lower() in (item_dict.get("title") or "").lower():
+                # Fuzzy relevance scoring
+                title = (item_dict.get("title") or "").lower()
+                query_lower = q.lower()
+                
+                # Exact match
+                if query_lower == title:
                     item_dict["_relevance_boost"] = 1.0
-                else:
+                # Title starts with query
+                elif title.startswith(query_lower):
+                    item_dict["_relevance_boost"] = 0.95
+                # Query is a substring of title
+                elif query_lower in title:
+                    item_dict["_relevance_boost"] = 0.85
+                # Words from query all appear in title
+                elif all(word in title.split() for word in query_lower.split() if len(word) > 1):
+                    item_dict["_relevance_boost"] = 0.7
+                # At least one word matches
+                elif any(word in title.split() for word in query_lower.split() if len(word) > 2):
                     item_dict["_relevance_boost"] = 0.5
+                else:
+                    item_dict["_relevance_boost"] = 0.3
+                
                 results[pid].append(item_dict)
 
                 # Prepare enrichment task for top 15 results per provider