Add backend parameter to detect_equilibration_window for optional numba dependencyupdate

[xiki-tempula]

Summary

• Added a backend parameter to detect_equilibration_window that allows choosing between "numba" (default) and "numpy" backends
• Made numba an optional dependency - the code now works without numba installed
• Added numba-accelerated parallel implementation for improved performance when numba is available
• Both backends produce identical results

Motivation

This change allows users who don't have numba installed (or don't want to use it) to still use detect_equilibration_window. The numpy backend provides a pure Python/numpy implementation that works without any additional dependencies.

Changes

• red/variance.py:
  • Made numba import optional with fallback
  • Added _optional_njit decorator that applies numba JIT only when available
  • Added _compute_variance_series_window_numpy - pure numpy implementation
  • Added _compute_variance_series_window_indexed - numba parallel implementation
  • Added backend parameter to get_variance_series_window
• red/sse.py: Added backend parameter to get_sse_series_window
• red/equilibration.py: Added backend parameter to detect_equilibration_window
• red/tests/test_equilibration.py: Added test_detect_equilibration_window_backends to verify both backends produce identical results

Usage

# Use numba backend (default, faster when numba is available)
idx, g, ess = red.detect_equilibration_window(data=series, times=times, method="min_sse", backend="numba")

# Use numpy backend (no numba dependency required)
idx, g, ess = red.detect_equilibration_window(data=series, times=times, method="min_sse", backend="numpy")