Migrate MLflow backend to PostgreSQL and fix .env loading

- Add python-dotenv loading in main.py so DATABASE_URL is read from .env
  before db.py module initializes
- Add MLFLOW_TRACKING_URI to .env.example pointing to PostgreSQL
- Add python-dotenv>=1.0.0 to pyproject.toml dependencies
- Initialize MLflow schema in candle_annotator PostgreSQL database

MLflow server now starts without filesystem deprecation warnings and
with full job execution support.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

services/ml/.env.example

@@ -1,3 +1,4 @@
 DATABASE_URL=postgresql://pg_user:pg_password@localhost:5432/candle_annotator
+MLFLOW_TRACKING_URI=postgresql://pg_user:pg_password@localhost:5432/candle_annotator
 # openssl rand -hex 32 add to main .env too
 API_KEY=ML_API_KEY

services/ml/app/main.py

@@ -17,6 +17,10 @@ from typing import Optional, Dict, Any, List
 from datetime import datetime, timezone
 import json
 
+# Load .env file before any module that reads environment variables
+from dotenv import load_dotenv
+load_dotenv(Path(__file__).parent.parent / ".env")
+
 import requests as http_requests
 
 from fastapi import FastAPI, HTTPException, Header, Depends, Security, status, Response

services/ml/pyproject.toml

@@ -19,6 +19,7 @@ dependencies = [
     "psycopg2-binary>=2.9.9",
     "pydantic>=2.5.0",
     "pydantic-settings>=2.1.0",
+    "python-dotenv>=1.0.0",
     "matplotlib>=3.8.2",
     "seaborn>=0.13.1",
 ]