Add ulcer index and Martin ratio to the returns accessor - #881

One definitional point on martin_ratio: the original Martin and McCann ulcer performance index is (annualized return minus the risk-free rate) divided by the ulcer index, the same way Sharpe uses excess return. The PR divides the raw annualized return, so the result only matches the textbook number when rf is 0, and it will read higher than other libraries' UPI for any non-zero rf. calmar_ratio here also has no rf, so staying consistent with it is defensible, but the docstring should say so ("annualized return, no risk-free adjustment") or the method could take a risk_free argument defaulting to 0 like sharpe_ratio/omega_ratio do. A small test with a non-zero rf against the pandas reference would pin whichever choice is made.

Thanks, good catch. martin_ratio and rolling_martin_ratio now take a risk_free argument (annualized, default 0) and the numerator is annualized return minus risk_free. The docstring says it defaults to the plain annualized return like calmar_ratio. I added a test with a non-zero rf against the pandas reference.

@polakowo this is ready for review. CI is green, and the earlier review point on risk_free is addressed. Small, self-contained addition next to calmar_ratio; happy to adjust naming or add the metrics to stats() if you prefer.