To sample points from isotropic spatial kernels
\(f_2(s) = f(||s||)\) such as siaf.powerlaw
on a
bounded domain (i.e., \(||s|| < \code{ub}\)), it is
convenient to switch to polar coordinates \((r,\theta)\),
which have a density proportional to
\(r f_2((r \cos(\theta), r \sin(\theta))) = r f(r)\)
(independent of the angle \(\theta\) due to isotropy).
The angle is thus simply drawn uniformly in \([0,2\pi)\), and
\(r\) can be sampled by the inversion method, where numeric root
finding is used for the quantiles (since the quantile function is not
available in closed form).