Fix problem with single power spectrum

[matteobachetti]

Resolve #663

Note for reviewers: I created a test reproducing exactly the MWE by @dhuppenkothen in #663, and a similar one for cross spectra. My first solution broke other tests, so I changed slightly the logic in fourier.py and crossspectrum.py to make everything work.
In particular, if a segment size is almost an integer number of light curve bins (less than 1% away), instead of approximating to the lower bin I approximate to the higher one. Not more than that, because a bin with a significantly lower flux could alter the statistical properties of the data. New function in utils.py for this

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (d47c95b) 96.37% compared to head (19d0c15) 96.27%.
Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #797      +/-   ##
==========================================
- Coverage   96.37%   96.27%   -0.11%     
==========================================
  Files          44       44              
  Lines        8824     8855      +31     
==========================================
+ Hits         8504     8525      +21     
- Misses        320      330      +10     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mgullik]

Hi @matteobachetti

I hope I understood what this PR is doing.
Isn't it better to check what is in the "last" bin that we are now considering if segment size is almost an integer number of light curve bins?
Since the segment size could be set as the GTI length, if the flux in the last bin is zero, it probably means that the GTI is set such as that bin shouldn't be included. Do you agree?

Moreover, I didn't understand what happended when there are multiple GTIs and segment size Is None. It seems that segment size becomes gti[-1, 1] - gti[0, 0] which would return an error (I hope) since the segment size is larger than all GTIs.

I would add tests with multiple GTIs in the case segment size is not set.

[dhuppenkothen]

LGTM overall. A couple of minor questions, but I don't think they should stop merging.

[matteobachetti]

@mgullik , @dhuppenkothen I think I've addressed all your comments.

Quick answers to @mgullik 's question:

Since the segment size could be set as the GTI length, if the flux in the last bin is zero, it probably means that the GTI is set such as that bin shouldn't be included. Do you agree?
Moreover, I didn't understand what happended when there are multiple GTIs and segment size Is None. It seems that segment size becomes gti[-1, 1] - gti[0, 0] which would return an error (I hope) since the segment size is larger than all GTIs.

This is a good point, and the code takes care of it. The GTIs always account for the bin size, so ideally they go from lc.time[0] - dt/2 to lc.time[-1] + dt / 2.