fix(classifier): fix LoRA classifier fallback and update E2E configs

This commit fixes a critical bug in LoRA intent classifier fallback logic
and updates E2E/deployment configs to use LoRA intent classification,
following the pattern established in PR #709 for PII detection.

Changes:
1. Fixed LoRA classifier fallback in classify_candle_bert_text
   - LoRA errors now fall through to Traditional BERT instead of returning
   - Ensures backward compatibility when LoRA models fail

2. Updated E2E and deployment configs to use LoRA intent classifier
   - e2e/profiles: ai-gateway, dynamic-config
   - deploy/kubernetes/aibrix: semantic-router-values
   - deploy/helm: semantic-router values + model download list

3. Fixed gofumpt formatting in lora_auto_detection_test.go

Fixes #724

Signed-off-by: $(git config user.name) <$(git config user.email)>
Signed-off-by: Yossi Ovadia <[email protected]>

Author: yossiovadia

candle-binding/src/ffi/classify.rs

@@ -709,8 +709,11 @@ pub extern "C" fn classify_candle_bert_text(text: *const c_char) -> Classificati
                 };
             }
             Err(e) => {
-                eprintln!("LoRA intent classifier failed: {}", e);
-                return default_result;
+                eprintln!(
+                    "LoRA intent classifier error: {}, falling back to Traditional BERT",
+                    e
+                );
+                // Don't return - fall through to Traditional BERT classifier
             }
         }
     }

deploy/helm/semantic-router/values.yaml

@@ -159,6 +159,8 @@ initContainer:
       repo: Qwen/Qwen3-Embedding-0.6B
     - name: all-MiniLM-L12-v2
       repo: sentence-transformers/all-MiniLM-L12-v2
+    - name: lora_intent_classifier_bert-base-uncased_model
+      repo: LLM-Semantic-Router/lora_intent_classifier_bert-base-uncased_model
     - name: category_classifier_modernbert-base_model
       repo: LLM-Semantic-Router/category_classifier_modernbert-base_model
     - name: pii_classifier_modernbert-base_model
@@ -272,11 +274,11 @@ config:
   # Classifier configuration
   classifier:
     category_model:
-      model_id: "models/category_classifier_modernbert-base_model"
-      use_modernbert: true
+      model_id: "models/lora_intent_classifier_bert-base-uncased_model"
+      use_modernbert: false  # Use LoRA intent classifier with auto-detection
       threshold: 0.6
       use_cpu: true
-      category_mapping_path: "models/category_classifier_modernbert-base_model/category_mapping.json"
+      category_mapping_path: "models/lora_intent_classifier_bert-base-uncased_model/category_mapping.json"
     pii_model:
       model_id: "models/pii_classifier_modernbert-base_presidio_token_model"
       use_modernbert: true

deploy/kubernetes/aibrix/semantic-router-values/values.yaml

@@ -431,11 +431,11 @@ config:
   # Classifier configuration
   classifier:
     category_model:
-      model_id: "models/category_classifier_modernbert-base_model"
-      use_modernbert: true
+      model_id: "models/lora_intent_classifier_bert-base-uncased_model"
+      use_modernbert: false  # Use LoRA intent classifier with auto-detection
       threshold: 0.6
       use_cpu: true
-      category_mapping_path: "models/category_classifier_modernbert-base_model/category_mapping.json"
+      category_mapping_path: "models/lora_intent_classifier_bert-base-uncased_model/category_mapping.json"
     pii_model:
       # Support both traditional (modernbert) and LoRA-based PII detection
       # When model_type is "auto", the system will auto-detect LoRA configuration

e2e/profiles/ai-gateway/values.yaml

@@ -461,11 +461,11 @@ config:
   # Classifier configuration
   classifier:
     category_model:
-      model_id: "models/category_classifier_modernbert-base_model"
-      use_modernbert: true
+      model_id: "models/lora_intent_classifier_bert-base-uncased_model"
+      use_modernbert: false  # Use LoRA intent classifier with auto-detection
       threshold: 0.6
       use_cpu: true
-      category_mapping_path: "models/category_classifier_modernbert-base_model/category_mapping.json"
+      category_mapping_path: "models/lora_intent_classifier_bert-base-uncased_model/category_mapping.json"
     pii_model:
       # Support both traditional (modernbert) and LoRA-based PII detection
       # When model_type is "auto", the system will auto-detect LoRA configuration

e2e/profiles/dynamic-config/values.yaml

@@ -42,11 +42,11 @@ config:
 
   classifier:
     category_model:
-      model_id: "models/category_classifier_modernbert-base_model"
-      use_modernbert: true
+      model_id: "models/lora_intent_classifier_bert-base-uncased_model"
+      use_modernbert: false  # Use LoRA intent classifier with auto-detection
       threshold: 0.6
       use_cpu: true
-      category_mapping_path: "models/category_classifier_modernbert-base_model/category_mapping.json"
+      category_mapping_path: "models/lora_intent_classifier_bert-base-uncased_model/category_mapping.json"
     pii_model:
       model_id: "models/lora_pii_detector_bert-base-uncased_model"
       use_modernbert: false  # Use LoRA PII model with auto-detection

src/semantic-router/pkg/classification/lora_auto_detection_test.go

@@ -34,7 +34,6 @@ func TestIntentClassificationLoRAAutoDetection(t *testing.T) {
 
 		// Try to initialize - should SUCCESS with LoRA auto-detection
 		err := initializer.Init(cfg.ModelID, cfg.UseCPU, numClasses)
-
 		if err != nil {
 			t.Errorf("Auto-detection failed: %v", err)
 			return