feat: 15 funciones datascience — estadística, DSP e IO de datos

12 funciones puras con implementación real:
Standardize, MinMaxScale, Clip, RollingWindow, ZipSlices, GroupBy,
Histogram, Pearson, Autocorrelation, FFT (Cooley-Tukey), DetectOutliers, Impute

3 funciones impuras (stubs):
LoadCSV, LoadParquet, FetchDataFrame

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

functions/datascience/fft.go

@@ -0,0 +1,61 @@
+package datascience
+
+import (
+	"math"
+	"math/cmplx"
+)
+
+// FFT calcula la Fast Fourier Transform usando el algoritmo Cooley-Tukey radix-2.
+// Si la longitud de data no es potencia de 2, se rellena con ceros (zero-padding).
+func FFT(data []float64) []complex128 {
+	n := len(data)
+	if n == 0 {
+		return []complex128{}
+	}
+
+	// Calcular la siguiente potencia de 2.
+	size := nextPow2(n)
+
+	// Convertir a complex128 con zero-padding.
+	x := make([]complex128, size)
+	for i := 0; i < n; i++ {
+		x[i] = complex(data[i], 0)
+	}
+
+	fftRecursive(x)
+	return x
+}
+
+// nextPow2 retorna la menor potencia de 2 >= n.
+func nextPow2(n int) int {
+	p := 1
+	for p < n {
+		p <<= 1
+	}
+	return p
+}
+
+// fftRecursive aplica Cooley-Tukey radix-2 DIT in-place.
+func fftRecursive(x []complex128) {
+	n := len(x)
+	if n <= 1 {
+		return
+	}
+
+	// Separar pares e impares.
+	even := make([]complex128, n/2)
+	odd := make([]complex128, n/2)
+	for i := 0; i < n/2; i++ {
+		even[i] = x[2*i]
+		odd[i] = x[2*i+1]
+	}
+
+	fftRecursive(even)
+	fftRecursive(odd)
+
+	for k := 0; k < n/2; k++ {
+		t := cmplx.Rect(1, -2*math.Pi*float64(k)/float64(n)) * odd[k]
+		x[k] = even[k] + t
+		x[k+n/2] = even[k] - t
+	}
+}