gabor crashes with ZeroDivisionError when bandwidth=0 #8301

What's happening

The gabor and gabor_kernel functions crash with a
ZeroDivisionError when bandwidth=0 is passed, instead of a clear
error message. The docstring doesn't say anywhere that bandwidth
has to be a positive number.

How to reproduce it

import numpy as np
from skimage.filters import gabor

image = np.zeros((10, 10))
gabor(image, frequency=0.1, bandwidth=0)

What actually happens

Why it fails

• Internally, gabor calls gabor_kernel, which calls a helper
  function _sigma_prefactor(bandwidth)
• That helper computes:
  1.0 / np.pi * sqrt(log(2) / 2.0) * (2.0**b + 1) / (2.0**b - 1)
• When bandwidth is 0, 2.0**0 equals 1, which makes the bottom
  part of that formula, (2.0**b - 1), come out to exactly 0
• This isn't a rare floating-point rounding issue — it's a guaranteed,
  exact division by zero for this one input
• Nowhere in gabor, gabor_kernel, or _sigma_prefactor is there
  any check that stops bandwidth=0 from reaching this line

Evidence this is a real, reproducible bug

• Reproduced it myself with the minimal script above — the traceback
  clearly shows the crash happening through gabor → gabor_kernel
  → _sigma_prefactor
• None of the existing doctest examples use bandwidth=0 — they only
  use the default (1) or 0.1
• None of the existing tests cover it either — the test for bandwidth
  only checks 1 and 0.5
• The docstring for the bandwidth parameter never mentions that it
  needs to be strictly positive

Suggested fix

Add a simple validation check near the start of gabor_kernel (which
gabor also goes through), something like:

if bandwidth <= 0:
    raise ValueError('bandwidth must be > 0')

Version info

scikit-image version: 0.25.2
numpy version: 2.1.3
Platform: Linux-6.6.122+-x86_64-with-glibc2.35