feat: relative width smoothing procedures

[jokasimr]

Adds functionality to scippneutron for smoothing 1D curves with kernels that have a width relative to the dimension coordinate.

As an example, in the figure below the step signal is smoothed by a gaussian kernel with width proportional to x:

The computed smoothed signal and the expected exact signal overlap.

[jokasimr]

The idea is that the scaled kernel has constant width on a grid that is the logarithm of the original grid, and when the kernel is constant width it can be computed efficiently using scipy.signal.convolve.

Here's a document describing the implementation in more detail:

relative_kernel_smoothing_math.pdf

[jokasimr]

@SimonHeybrock Do you think this functionality should go here in Scippneutron or in Scipp?

It's mainly useful for analysis work where you have an idealized model and want to apply some smoothing from resolution effects to obtain something that is comparable to the measurement data.