bug#21906: julian-day->date negative input breakage

[Zefram]

scheme@(guile-user)> (use-modules (srfi srfi-19))
scheme@(guile-user)> (julian-day->date 0 0)
$1 = #<date nanosecond: 0 second: 0 minute: 0 hour: 12 day: 24 month: 11 year: 
-4714 zone-offset: 0>
scheme@(guile-user)> (julian-day->date -1 0)
$2 = #<date nanosecond: 0 second: 0 minute: 0 hour: -12 day: 24 month: 11 year: 
-4714 zone-offset: 0>
scheme@(guile-user)> (julian-day->date -100000 0)
$3 = #<date nanosecond: 0 second: 0 minute: 0 hour: -12 day: -19 month: 3 year: 
-4987 zone-offset: 0>
scheme@(guile-user)> (julian-day->date -10000000 0)
$4 = #<date nanosecond: 0 second: 0 minute: 0 hour: -12 day: -30 month: 0 year: 
-32092 zone-offset: 0>

Observe the various erroneous field values: negative hour, negative
day-of-month, zero month.  These occur in general for various negative
JD inputs.  Not only should the conversion not produce these kinds of
values, the date structure type probably ought to reject them if they
get that far.

-zefram