Fix for bug #21591 (https://gna.org/bugs/?21591), the complete failure of the NOE analysis.
This bug was reported by Martin Ballaschk (https://gna.org/users/mab).
The issue was introduced in the fix for bug #21562 (https://gna.org/bugs/?21562). The problem is
that the overfit_deselect() method was deselecting all spins with two data points or less rather
than one or less.
