Add US stock 5-min prediction script using yfinance

[akommos88-source]

What changed

Adds predict_us_stock.py — a self-contained script for running Kronos inference on US equities using live 5-minute OHLCV data fetched from Yahoo Finance.

Why

The existing examples in examples/ rely on pre-downloaded CSV files from Chinese markets (XSHG). This script makes it easy for users with no existing data pipeline to run Kronos on any US stock ticker out of the box.

What it does

1. Fetches data — pulls up to 30 days of 5-min bars via yfinance, strips pre/post market hours (09:30–16:00 ET), and selects the most recent lookback + pred_len candles
2. Loads models — downloads KronosTokenizer + Kronos from HuggingFace Hub on first run (cached after); supports local paths via --tokenizer / --model flags
3. Runs inference — uses KronosPredictor.predict() with configurable temperature, top-p, and Monte Carlo sample count (averaged for smoother output)
4. Outputs results — prints predicted OHLCV table + directional summary (UP/DOWN % vs last historical close), saves a two-panel close + volume PNG chart

Usage

pip install yfinance  # only additional dependency beyond requirements.txt
python predict_us_stock.py --ticker AAPL --lookback 200 --pred_len 20
python predict_us_stock.py --ticker TSLA --lookback 300 --pred_len 10 --samples 10 --temp 0.8

Notes for reviewer

• The amount column required by KronosPredictor is not provided by yfinance; the predictor handles this gracefully (fills with zeros) when the column is absent — no changes to model code needed.
• No existing files were modified; this is a pure addition.
• The pretrained models were trained primarily on global/Asian exchange data, so directional accuracy on US stocks will be limited without fine-tuning via finetune_csv/.