feat: Add comprehensive risk metrics to simulation results

[Copilot]

Description

The simulation returned only basic metrics (final value, return, buy-and-hold, trade count), insufficient to evaluate risk-adjusted strategy quality. This adds 11 standard risk metrics computed from the portfolio history and trade log.

Changes in this pull request

New: src/simulation/risk_metrics.py

• RiskMetrics dataclass with 11 fields; any field is None when there's insufficient data
• compute_risk_metrics(portfolio_history, trades, total_return) computing:
  • Sharpe / Sortino — annualised, risk-free rate = 0; Sortino uses downside std dev only
  • Max drawdown — peak-to-trough fraction; drawdown duration — consecutive days below prior peak
  • Calmar ratio — annualised return / |max drawdown|
  • Win rate, profit factor, avg win/loss ratio — derived from pnl_pct per trade
  • Annualised volatility — daily return std dev × √252
  • VaR (95%) — 5th percentile of daily portfolio returns
  • Beta to benchmark — cov(strategy returns, price returns) / var(price returns)

Modified: src/simulation/trading_simulator.py

• simulate() calls compute_risk_metrics() at the end; result exposed under "risk_metrics" key

Modified: src/simulation/__init__.py

• Exports RiskMetrics and compute_risk_metrics

New: tests/simulation/test_risk_metrics.py

• 32 tests: edge cases (empty/single-row → all None), known-value assertions, sign checks, and integration tests on simulate() output

result = sim.simulate(predictions, actual_returns, prices=prices, dates=dates)
rm = result["risk_metrics"]  # RiskMetrics instance
print(rm.sharpe_ratio, rm.max_drawdown, rm.win_rate)