Add nan fail test

[xywei]

If the expansion basis is singular at the origin, and a source point lies at the expansion center, p2e fails with NaNs.
The added test shows that this is indeed an issue with Y2DMultipoleExpansion.

cf.

• Fourier-Bessel (possibly other) multipoles fail to form (produce NaNs) if there's a source coincident with a center #11
• https://gitlab.tiker.net/xywei/volumential/-/issues/43

[inducer]

Thanks!

cc @isuruf

[isuruf]

HelmholtzConformingVolumeTaylorMultipoleExpansion works? H2DMultipoleExpansion will not be the default Helmholtz expansion in the near future.

[xywei]

HelmholtzConformingVolumeTaylorMultipoleExpansion works? H2DMultipoleExpansion will not be the default Helmholtz expansion in the near future.

I think that Taylor expansions are okay because they are smooth. The issue pertains only to basis functions that are singular (like Bessel functions of the second kind).

[inducer]

I think that Taylor expansions are okay because they are smooth.

Taylor multipoles aren't smooth though. (The multipole basis consists of derivatives of the kernel.)