feat(ai-services): AVM v2 residential — expanded features, training pipeline, model versioning

Add neighborhood_score, developer_reputation, floor_level, direction premiums
to the multi-model ensemble. Implement real Optuna-based training pipeline
for XGBoost/LightGBM/CatBoost with grouped train/val/test splits. Add
file-based model registry with rollback and list-versions endpoints.
23 Python tests covering all new features.

Co-Authored-By: Paperclip <noreply@paperclip.ing>

libs/ai-services/app/models/avm_v2.py

@@ -29,10 +29,28 @@ class AVMv2PredictRequest(BaseModel):
         0.0, ge=0, le=1, description="Flood zone risk score (0=safe, 1=high risk)"
     )
 
+    # ── Neighborhood features ─────────────────────────────
+    neighborhood_score: float = Field(
+        0.5, ge=0, le=1,
+        description="Overall neighborhood quality score (0-1, aggregated from safety, amenities, walkability)",
+    )
+
     # ── Physical features ──────────────────────────────────
     property_type: str = Field(..., description="e.g. apartment, house, villa, land")
     area_m2: float = Field(..., gt=0, description="Property area in m²")
     rooms: int = Field(0, ge=0, description="Total rooms (bedrooms)")
+    floor_level: int = Field(
+        0, ge=0,
+        description="Floor level (0=ground or N/A, relevant for apartments/penthouses)",
+    )
+    total_floors: int = Field(
+        0, ge=0,
+        description="Total floors in the building (0=N/A)",
+    )
+    direction: str = Field(
+        "unknown",
+        description="Facing direction: north, south, east, west, northeast, northwest, southeast, southwest, unknown",
+    )
     floor_ratio: float = Field(
         1.0, gt=0, description="Total floor area / land area ratio"
     )
@@ -41,6 +59,10 @@ class AVMv2PredictRequest(BaseModel):
     has_parking: bool = Field(False, description="Property has dedicated parking")
     has_pool: bool = Field(False, description="Property has swimming pool")
     has_legal_paper: bool = Field(True, description="Has sổ đỏ/sổ hồng")
+    developer_reputation: float = Field(
+        0.5, ge=0, le=1,
+        description="Project developer reputation score (0-1, based on past projects, delivery record)",
+    )
 
     # ── Market features ────────────────────────────────────
     avg_price_district_3m_vnd_m2: float = Field(
@@ -185,6 +207,12 @@ class AVMv2ModelInfo(BaseModel):
     ab_test_traffic_pct: float = Field(0.0, ge=0, le=1)
 
 
+class AVMv2RollbackRequest(BaseModel):
+    """Request to rollback to a specific model version."""
+
+    target_version: str = Field(..., min_length=1, description="Model version to roll back to")
+
+
 class AVMv1Summary(BaseModel):
     """Compact summary of a v1 prediction for comparison."""
 
@@ -220,13 +248,18 @@ class ABComparisonRequest(BaseModel):
     frontage: float = Field(0.0, ge=0)
     has_legal_paper: bool = Field(True)
     # v2-specific features (optional, defaults applied)
+    neighborhood_score: float = Field(0.5, ge=0, le=1)
     distance_to_cbd_km: float = Field(0.0, ge=0)
     distance_to_metro_km: float = Field(0.0, ge=0)
     flood_zone_risk: float = Field(0.0, ge=0, le=1)
     building_age_years: int = Field(0, ge=0)
+    floor_level: int = Field(0, ge=0)
+    total_floors: int = Field(0, ge=0)
+    direction: str = Field("unknown")
     has_elevator: bool = Field(False)
     has_parking: bool = Field(False)
     has_pool: bool = Field(False)
+    developer_reputation: float = Field(0.5, ge=0, le=1)
     renovation_score: float = Field(0.5, ge=0, le=1)
     view_quality: float = Field(0.5, ge=0, le=1)
     interior_quality: float = Field(0.5, ge=0, le=1)