Confused about noise power spectral density measurement

[Malamala]

Hello! My question was started from this paper. The topic of the paper is related to linewidth measurements, but my question is even before that, related to the way the noise power spectral density (PSD) spectrum shown in Fig. 1 is obtained. In Fig. 2 they show their experimental setup, which is based on a self-heterodyne measurement. The time delay between the 2 arms is 100 ns. I am confused about the lower limit of the spectrum in Fig. 1. Naively, I would expect to not be able to see any noise faster than 1/(100 ns) = 10 MHz i.e. the 2 arms would appear coherent for noise slower than that. Yet, in Fig. 1 it appears like the 10 MHz is actually an upper, not a lower limit for the noise. On the right column of Fig. 2 they also claim "Therefore, the interferometric phase signal allows to extract the full PSD of laser frequency noise down to frequencies limited by the overall integration time of the interferometric phase signal". I don't understand why measuring for longer would give me information about slower noise variation. And if that is the case, why do experiments measuring noise PSD or linewidths in laser actually use 10's or even 100 km long fiber differences, if you can reach down to 1 Hz just by doing the self-heterodyne for 1 second?