feat(ai-services): add Python FastAPI AI/ML services container

Create libs/ai-services/ with FastAPI app providing:
- POST /avm/predict — XGBoost-backed property price prediction (heuristic fallback)
- POST /avm/extract-features — Vietnamese NLP feature extraction from listing text
- POST /moderation/check — content moderation with rule-based flagging
- GET /health — health check endpoint

Includes Dockerfile (Python 3.12), docker-compose integration, Pydantic models,
and 9 passing tests covering all endpoints.

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

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

@@ -0,0 +1,48 @@
+from pydantic import BaseModel, Field
+
+
+class AVMPredictRequest(BaseModel):
+    area: float = Field(..., gt=0, description="Property area in m²")
+    district: str = Field(..., min_length=1, description="District name")
+    city: str = Field(..., min_length=1, description="City name")
+    property_type: str = Field(..., description="e.g. apartment, house, land")
+    bedrooms: int = Field(0, ge=0)
+    bathrooms: int = Field(0, ge=0)
+    floors: int = Field(0, ge=0)
+    frontage: float = Field(0.0, ge=0, description="Frontage width in meters")
+    road_width: float = Field(0.0, ge=0, description="Adjacent road width in meters")
+    year_built: int | None = Field(None, description="Year the property was built")
+    has_legal_paper: bool = Field(True, description="Whether property has sổ đỏ/sổ hồng")
+
+
+class AVMPredictResponse(BaseModel):
+    estimated_price_vnd: float = Field(..., description="Estimated price in VND")
+    confidence: float = Field(..., ge=0, le=1, description="Prediction confidence score")
+    price_per_m2: float = Field(..., description="Price per m² in VND")
+    price_range_low: float = Field(..., description="Lower bound estimate in VND")
+    price_range_high: float = Field(..., description="Upper bound estimate in VND")
+
+
+class FeatureExtractRequest(BaseModel):
+    text: str = Field(..., min_length=1, description="Vietnamese property listing text")
+
+
+class ExtractedFeatures(BaseModel):
+    area: float | None = None
+    district: str | None = None
+    city: str | None = None
+    property_type: str | None = None
+    bedrooms: int | None = None
+    bathrooms: int | None = None
+    floors: int | None = None
+    frontage: float | None = None
+    road_width: float | None = None
+    price_mentioned: float | None = None
+    has_legal_paper: bool | None = None
+    address_raw: str | None = None
+
+
+class FeatureExtractResponse(BaseModel):
+    features: ExtractedFeatures
+    tokens: list[str] = Field(default_factory=list, description="Tokenized words")
+    entities: list[dict] = Field(default_factory=list, description="Named entities found")

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

@@ -0,0 +1,20 @@
+from pydantic import BaseModel, Field
+
+
+class ModerationRequest(BaseModel):
+    text: str = Field(..., min_length=1, description="Text content to moderate")
+    context: str = Field("listing", description="Context: listing, comment, profile")
+
+
+class ModerationFlag(BaseModel):
+    category: str
+    severity: str = Field(..., description="low, medium, high")
+    matched_text: str
+    reason: str
+
+
+class ModerationResponse(BaseModel):
+    is_flagged: bool
+    score: float = Field(..., ge=0, le=1, description="Overall risk score")
+    flags: list[ModerationFlag] = Field(default_factory=list)
+    cleaned_text: str | None = Field(None, description="Text with flagged content redacted")