Spring starts then. Not only is the sun moving from a negative declination to a positive one, but the Right Ascension is moving from just under 24h back to 0h. All those zeroes can allow minor math errors to become hugely important.

Without installing the modules myself, it looks like there may be a math error in the multiple (unnecessary) degree -> radian -> degree conversions leading to one too many factors of &pi/180 relating to the altitude term. If your problem goes away when you use the altitude of zero, that would confirm a problem in the _sunrise_sunset routine.

Assuming it is true for other years, I encourage you to report this bug to the author even if the altitude suggestion is incorrect.