Include padding in FourierFilters

Currently, FourierFilter from hotopy.image._filter applies a filter in FFT space (surprise), which assumes periodic boundary conditions. A common use case for GaussianBlur and GaussianBandpass would require padding to reduce boundary artifacts like

from hotopy.utils import Padder
from hotopy.image.filter import GaussianBlur
import numpy as np

data = np.random.random((128, 128)) + np.linspace(0, 1, 128)[None,:]
sigma = 3

npad = 2 * sigma  # npad = 0 produces edge artifacts
padder = Padder(npad, mode="edge")
padded_shape = tuple(s + 2 * npad for s in data.shape)
gaussfilt = GaussianBlur(padded_shape, sigma)
filtered_data = padder.inv(gaussfilt(padder(data)))

I would prefer the option to include padding in FourierFilter already. Something like:

from hotopy.image.filter import GaussianBlur
import numpy as np

data = np.random.random((128, 128)) + np.linspace(0, 1, 128)[None,:]
sigma = 3

gaussfilt = GaussianBlur(data.shape, sigma, npad=2 * sigma)
filtered_data = gaussfilt(data)

It would make sense to let FourierFilter.__init__ accept npad or maybe even a Padder instance and the inheriting classes calling super().__init__.

Maybe there is even some smart and fast FFT implicit padding way to do the padding, would need mode="edge" padding in dataspace though... for now that sounds like premature optimization to me.