Fix a 20+ year bug by using an appropriate constant for
the transition from one asymptotic approximation to another
for the zeroth order Bessel and Neumann functions.

Reviewed by: bde