Create some post-run result tests

[mcdigman]

We should have some more sanity checks after the run completes (other than just checking for normality of the residuals). Some ideas:

• check the ratios of the total detectable and undetectable snrs make sense relative to the total unresolvable power
• check the A and E channel snrs make sense relative to each other (e.g. total snr should be comparable, at least for observing times that are multiples of a year)
• check whether A and E channel noise spectrum shapes are similar
• check the evolution of the total power in the different components of the background makes sense; i.e. the faint and bright components should be either ~monotonically increasing or basically constant, and undecided power should be ~monotonically decreasing
• check the evolution of the noise power in the upper and lower components appears to converge in a sensible way
• check whether the final level of smoothing seems appropriate; e.g. how susceptible we are to binaries interfering with their own power
• check no binaries disappear in any accounting
• check T channel snr and signals are << A/E channel (if it was computed at all)
• check whether the power in the fourier components of the cyclostationary model are similar between the A and E channels, even if the angles are not.

Separate from #5, although reinforcing some of the same goals.