feat(fitness): add RiskAdjustedReturn fitness function (closes #21)

[nandanadileep]

Summary

Closes #21.

Adds a RiskAdjustedReturnFitness builtin that scores strategies using:

score = annual_return / (1 + |max_drawdown|)

This naturally penalises high-drawdown strategies even when their raw return is higher.

• Handles max_drawdown in either sign convention (stored negative or positive) via abs()
• Zero-trade results receive a -1.0 penalty score
• Auto-discovered by FitnessRegistry.auto_discover() via FITNESS_CLASSES

Files changed

File	Description
stratevo/fitness_plugins/builtin/risk_adjusted_return.py	New fitness class
tests/test_fitness_risk_adjusted_return.py	11 tests covering formula, ranking, edge cases, registry

Test plan

• test_formula_standard — verifies annual_return / (1 + dd_abs) arithmetic
• test_formula_positive_drawdown_convention — abs() handles positive dd values
• test_drawdown_penalises_high_return — 30%/90%DD loses to 20%/10%DD
• test_same_return_lower_drawdown_wins — same return, lower DD scores higher
• test_zero_trades_penalty — returns -1.0
• test_zero_drawdown — denominator = 1, score = annual_return
• test_negative_return — negative returns produce negative scores
• test_large_drawdown — formula still finite at 80% drawdown
• test_auto_discover_registers_it — registry finds the class automatically

Note on issue example numbers: The issue states "20% return / 10% DD scores higher than 30% return / 25% DD" — mathematically this does not hold (0.182 < 0.240). The tests instead verify the actual behaviour of the formula: higher drawdown penalises even high-return strategies when the drawdown is sufficiently large, and equal-return strategies are correctly ranked by drawdown.

[NeuZhou]

Excellent PR. Clean code, thorough tests, and the docstring honestly acknowledges the math correction from the issue spec. CI green.

✅ Highlights

• Formula is simple and correct: \�nnual_return / (1 + |max_drawdown|)\
• Handles both sign conventions for drawdown via \�bs()\ — important since different callers may store it differently
• Zero-trade penalty of -1.0\ is a good choice
• 11 tests covering formula, ranking, edge cases, and registry integration
• Properly uses \FITNESS_CLASSES\ for auto-discovery
• Honest note about the issue example math being off — good attention to detail

Minor suggestion (non-blocking)

The docstring comment about the issue spec's math error is nice for developers. Consider trimming it slightly since it reads like a review discussion — maybe just \Note: abs() normalizes sign convention.\

Approved ✅