This test reproduces the runaway problem described in github issue #240:
https://github.com/LinuxCNC/linuxcnc/issues/240
The input is a sine wave (sampled from siggen), amplitude 100, that is
abruptly cut off a -42, on the way down on the first cycle.