Inexactness in julian_day() function.

If we compute Julian Day Number for 13th August 2020 using julian_day() we get **2459075**.
But actual Julian Day number is **2459074.5**.

For 22nd April 1980 using julian_day() we get **2444352**.
Actual Julian Day number is **2444351.5**.

For 7th September 1999 using julian_day() we get **2451429**.
Actual Julian Day number is **2451428.5**.

There is always a difference of **0.5** in actual and what julian_day() number is returning.

**Verified from**
| | |
| ------ | ------ |
| Keisan Online Calculator | https://keisan.casio.com/exec/system/1227779487 |
| Neoprogrammics | http://neoprogrammics.com/sidereal_time_calculator/index.php |

I am working on Boost.Astronomy and Julian Day number of a day is required when calculating Greenwich Sidereal Time.

Start by locating the Boost.Date_Time julian_day() entry point and inspect how its return value is defined relative to Julian Day Number and Julian Date. Reproduce the three dates from the report, then add regression coverage that makes the intended half-day behavior explicit.

Vom Indexierungsmodell aus dem Issue-Text verfasst.