Fix deployment: create missing Flask app, save preprocessor, improve KMeans efficiency

[Copilot]

The app could not be deployed because app/main.py didn't exist, the app/ directory was tracked as a broken git submodule, the fitted preprocessor was never persisted (making inference impossible), and KMeans lacked an explicit n_init causing sklearn deprecation warnings.

Changes

• app/main.py (new): Flask app with GET / health-check and POST /predict (CSV upload → segment labels). Model and preprocessor loaded lazily on first request.
• app/__init__.py / src/__init__.py (new): Package init files so imports resolve correctly.
• generate_model.py: Capture and persist the fitted preprocessor to models/preprocessor.pkl — was previously discarded (X, _ = ...), making it impossible to transform new data consistently at inference time.
• src/clustering.py: Add n_init=10 explicitly to KMeans to silence sklearn FutureWarning and lock in consistent behavior.
• run_app.sh: Replace export FLASK_APP=... && flask run with python app/main.py.
• Git: Removed broken submodule entry for app/ (no .gitmodules existed); re-added as regular tracked files.

# POST /predict — upload CSV, get back segment labels
# app/main.py
@app.route("/predict", methods=["POST"])
def predict():
    df = pd.read_csv(io.StringIO(file.stream.read().decode("utf-8")))
    _load_artifacts()                          # lazy-loads model + preprocessor
    df = engineer_features(df)
    X = _preprocessor.transform(df)           # uses saved preprocessor
    labels = _model.predict(X)
    return jsonify({"segments": labels.tolist()})

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.