DST transitions cause APScheduler to incorrectly calculate next trigger times

[rodrigoformi-zh]

Description
DST transitions cause APScheduler to incorrectly calculate next trigger times.

Given:
a cron trigger of month='', day='', day_of_week='0-4', hour='0', minute='0',
a last trigger time of None
a current time of pd.Timestamp(2021, 3, 12, 0, 0,0,1,0,pytz.timezone('US/Central'))

APScheduler returns '2021-03-16T00:00:00-05:00' for the cron_trigger's get_next_fire_time() call.

NOTE: there is a DST transition CST -> CDT at 2AM on 14 March.

from apscheduler.triggers.cron import CronTrigger
import pytz
import pandas as pd

cron_trigger = CronTrigger.from_crontab('0 0 * * 0-4', pytz.timezone('US/Central'))
now = pd.Timestamp(2021, 3, 12, 0, 0,0,1,0,pytz.timezone('US/Central'))
print(cron_trigger.get_next_fire_time(None,now).isoformat())

Expected behavior
To return '2021-03-15T00:00:00-05:00' not '2021-03-16T00:00:00-05:00'

Additional context
The APScheduler library determines the next fire time by going through each part of the expression and finding the next viable value iteratively until all parts of the expression are satisfied. When the day is incremented, 24 hours are added. This causes the time to go forward 1 day and 1 hour as DST causes a day to only contain 23 hours.

It seems that the _increment_field_value function needs to be corrected to accommodate DST changes
NOTE: the same behavior doesn't happen on CDT->CST transitions

[agronholm]

Seems like this is an unfortunate corner case that nobody else has come across these last 12 years.

If I adjust the code to make it work with this particular case, two other cases start failing in the test suite.

There are 4 test cases in the test suite that should have ensured correct operation across DST boundaries (here).