Bug in PyCBC reproducible noise: Discontinuities present over frame boundaries

[spxiwh]

Details from @ArthurTolley:

We are trying to use pycbc_condition_strain to make a long stretch of simulated data with short frame files (to test PyCBC Live like operation). This saw a lot of glitches, Arthur produced the following minimal example:

fake_strain=aLIGODesignSensitivityT1800044
ifo=H1
duration=32
out_path="./${ifo}-SIMULATED_STRAIN-{start}-{duration}.gwf"

for chunk in {1..2}; do
    gps_start_time=$((1200000000+(${chunk}-1)*64))
    gps_end_time=$((1200000000+${chunk}*64))

    pycbc_condition_strain \
        --fake-strain ${fake_strain} \
        --fake-strain-seed 1234 \
        --output-strain-file ${out_path} \
        --gps-start-time ${gps_start_time} \
        --gps-end-time ${gps_end_time} \
        --sample-rate 2048 \
        --frame-duration ${duration} \
        --channel-name $ifo:SIMULATED_STRAIN \

done

which produces two data chunks with a discontinuous boundary (Q-scan over boundary):

I can't see anything wrong here. Arthur says if the PSD is changed to aLIGOMidLowSensitivityP1200087 the data is good, but we shouldn't have "bad" and "good" PSDs in this context.