Ignore imaginary frequencies with low absolute values

[danielhollas]

The imaginary frequencies with absolute values below 50 cm^-1 likely come from numerical noise, so it's probably safe to ignore them instead of killing the calculation (but we should issue a warning in the logs at least). Since we typically strip frequencies below 100cm^-1 that seems like a correct things to do. But some experimentation and discussion is needed.

See calculation pk=449761.