Implement Skewness Calculation with Edge Case Handling

[ANANDHA007]

This PR adds a Skewness function for float64 datasets, calculating the third central moment normalized by standard deviation cubed. It handles zero-variance datasets and validates minimum input length. Unit tests cover symmetric, right-skewed, and left-skewed distributions, complementing the library’s existing descriptive statistics.

[ANANDHA007]

Hi @montanaflynn added the functional logic to calculate the Skewness

[montanaflynn]

Thanks for adding skewness — great addition to the library! I pushed a commit with some adjustments to align with the existing codebase conventions:

Formula fix:

• The original formula computed sumOfCubes / stdDevCubed, but sumOfCubes is the raw sum Σ(xi - μ)³ while stdDevCubed was derived from variance = Σ(xi - μ)² / n (already divided by n). This made the result off by a factor of n. Fixed to (sumOfCubes / n) / stdDevCubed.

Added population/sample variants:

• Skewness → alias for PopulationSkewness (matches scipy.stats.skew(bias=True))
• PopulationSkewness → third central moment / σ³
• SampleSkewness → adjusted Fisher-Pearson coefficient with bias correction: g1 * sqrt(n*(n-1)) / (n-2) (matches scipy.stats.skew(bias=False))

This follows the same pattern as Variance/PopulationVariance/SampleVariance and StandardDeviation/StandardDeviationPopulation/StandardDeviationSample.

Convention fixes:

• Use ErrEmptyInput instead of errors.New(...) — matches the package's error pattern
• Return math.NaN() on error instead of 0.0 — matches every other function in the library
• Unexported Differences() and PowerofN() helpers and inlined the logic — keeps the public API clean
• Switched tests to package stats_test and table-driven format with exact reference values verified against scipy
• Added doc comments on exported functions

All tests pass against scipy reference values for both population and sample skewness across 5 datasets.