Your observation of what the doc says is correct. However, opinions may vary on whether it represents a typo or a bug. It reflects what the functions do. And what they do is arguably in line with <u>[Postel's law](https://en.wikipedia.org/wiki/Robustness_principle)</u>, which urges leniency on input and rigor on output.
As an example for discussion consider this session screen-scrape:<code>
sqlite> select datetime( julianday('2000-01-01T24:59') );
2000-01-02 00:59:00
</code>. The result is sensible, and allows somebody who may be less than utterly familiar with the 24-hour time format to speak of the earliest morning hour without mention of the zero hour. (The fact that it's in the next day is another issue; the result is logical.)
There will be no mention of hour 24 in 24-hour output from the datetime/time functions. And whether somebody exploits the input leniency is up to them.