erf(inf)



As mentioned a few days ago, maxima automatically simplifies erf(inf) to
1.  And also atan(inf) to %pi/2.  This seems intentional, but can lead
to interesting things like

        (atan(inf)-%pi/2)*inf => 0
but
        limit(atan(x)-%pi/2)*x,x,inf) => -1

as Stavros mentioned.

Therefore, I think we should change this behavior so that erf(inf) is
erf(inf) and not 1.

But since this seems intentional, I'm soliciting opinions on this.  This
also affects erfc, erfi, gamma_incomplete, fresnel_s and fresnel_c.
There might be others as well.

Ray