[FEA] Provide GPU utilization metric for benchmarks about cucim HOT 1 OPEN

The initial thought of trying to using pynvml turned out to not work. Instead, I profiled the following with nsight compute:

import cupy as cp
from cucim.skimage.filters import gaussian, gabor
a = cp.random.random((30_000, 30_000))
a[1, 1] = 1
a[1000, 1] = 1
a[3000, 1] = 1

gabor(a, frequency=0.4)

I don't think compute/gpu utilization is the best metric here as the above does not really inform how well gabor performs for marketing materials. Engineers might be interested here and explore why this may or may not be efficient but I don't think this on its own conveys much to a general image scientist

from cucim.