performance update raw python implementation

[wouterverweirder]

I've been able to get a 20x performance update for the raw python implementation. (and 40x when using a caching mechanism).

1. Added Numba JIT compilation (@jit(nopython=True)) to:
   • thinningZSIteration() - The most critical bottleneck with nested loops
   • chunkToFrags() - Another function with nested loops
2. Fixed notEmpty() function - Changed from np.sum(im) > 0 to np.any(im[y:y+h, x:x+w]) which:
   • Properly slices the image region (the original version summed the entire image!)
   • Uses np.any() which is faster and more appropriate for boolean checks
3. Replaced nested loops with NumPy slicing in traceSkeleton():
   • Horizontal seam search: np.sum(im[i-1:i+1, x:x+w]) instead of nested loop
   • Vertical seam search: np.sum(im[y:y+h, j-1:j+1]) instead of nested loop