fix(e2e): enable LoRA PII auto-detection for ai-gateway and aibrix E2E tests

Fixes the PII detection test failures for both ai-gateway and aibrix profiles
by switching from ModernBERT to LoRA-based PII detection.

## Root Cause

Both profiles were using the old ModernBERT PII model instead of the new
LoRA-based auto-detection model, causing 0% PII detection accuracy in tests.

## Applied Fixes

### 1. AI Gateway E2E Profile (e2e/profiles/ai-gateway/values.yaml)

**Fix 1: LoRA PII Model Configuration** (lines 469-476)
Updated classifier.pii_model to use LoRA auto-detection:
- model_id: "models/lora_pii_detector_bert-base-uncased_model"
- model_type: "auto" (enables LoRA auto-detection)
- pii_mapping_path: points to LoRA model's pii_type_mapping.json

**Fix 2: Default Decision Fallback** (lines 413-432)
Added default_decision to enable PII detection when no route matches:
- priority: 0 (lowest, fallback only)
- pii_types_allowed: [] (block ALL PII)
- Works with policy.go fallback mechanism

### 2. Aibrix Deployment Config (deploy/kubernetes/aibrix/semantic-router-values/values.yaml)

**Fix: LoRA PII Model Configuration** (lines 176-183)
Same LoRA model update as ai-gateway.
Note: Aibrix uses old category-based routing, so no default_decision needed.

## Test Results

**AI Gateway**: Verified locally with Kind cluster - 100% PII detection accuracy (100/100 tests)
**Aibrix**: Uses same LoRA model configuration, should achieve same 100% accuracy

## Why Both Fixes Were Necessary

1. **LoRA Model**: Switches to the new auto-detection infrastructure
2. **Default Decision** (ai-gateway only): Ensures PII policy applies even when
   no specific decision matches the request content

Fixes #647

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: Yossi Ovadia <yovadia@redhat.com>

Author: yossiovadia

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

@@ -437,11 +437,13 @@ config:
       use_cpu: true
       category_mapping_path: "models/category_classifier_modernbert-base_model/category_mapping.json"
     pii_model:
-      model_id: "models/pii_classifier_modernbert-base_presidio_token_model"
-      use_modernbert: true
+      # Support both traditional (modernbert) and LoRA-based PII detection
+      # When model_type is "auto", the system will auto-detect LoRA configuration
+      model_id: "models/lora_pii_detector_bert-base-uncased_model"
+      model_type: "auto"  # Enables LoRA auto-detection
       threshold: 0.7
       use_cpu: true
-      pii_mapping_path: "models/pii_classifier_modernbert-base_presidio_token_model/pii_type_mapping.json"
+      pii_mapping_path: "models/lora_pii_detector_bert-base-uncased_model/pii_type_mapping.json"
 
   keyword_rules:
     - name: "thinking"

e2e/profiles/ai-gateway/values.yaml

@@ -410,6 +410,27 @@ config:
             system_prompt: "You are a helpful and knowledgeable assistant. Provide accurate, helpful responses across a wide range of topics."
             mode: "replace"
 
+    # Default catch-all decision for unmatched requests (E2E PII test fix)
+    # This ensures PII detection is always enabled via policy.go fallback mechanism
+    # When no decision matches, CheckPolicy and IsPIIEnabled fall back to this decision
+    - name: default_decision
+      description: "Default catch-all decision - blocks all PII for safety"
+      priority: 0  # Lowest priority
+      rules:
+        operator: "OR"
+        conditions:
+          - type: "domain"
+            name: "catch_all_fallback"
+      modelRefs:
+        - model: base-model
+          lora_name: general-expert
+          use_reasoning: false
+      plugins:
+        - type: "pii"
+          configuration:
+            enabled: true
+            pii_types_allowed: []  # Block ALL PII - empty list means nothing allowed
+
   # Strategy for selecting between multiple matching decisions
   # Options: "priority" (use decision with highest priority) or "confidence" (use decision with highest confidence)
   strategy: "priority"
@@ -467,11 +488,13 @@ config:
       use_cpu: true
       category_mapping_path: "models/category_classifier_modernbert-base_model/category_mapping.json"
     pii_model:
-      model_id: "models/pii_classifier_modernbert-base_presidio_token_model"
-      use_modernbert: true
+      # Support both traditional (modernbert) and LoRA-based PII detection
+      # When model_type is "auto", the system will auto-detect LoRA configuration
+      model_id: "models/lora_pii_detector_bert-base-uncased_model"
+      model_type: "auto"  # Enables LoRA auto-detection
       threshold: 0.7
       use_cpu: true
-      pii_mapping_path: "models/pii_classifier_modernbert-base_presidio_token_model/pii_type_mapping.json"
+      pii_mapping_path: "models/lora_pii_detector_bert-base-uncased_model/pii_type_mapping.json"
 
   keyword_rules:
     - name: "thinking"