test: reproduce #74 — rrule skips the occurrence at dtstart

[sibson]

Adds a failing regression test for #74, reproduced on main (2f96d0a).

The mechanism

An rrule built without an explicit dtstart defaults it to the creation time
(redbeat/schedules.py:53), and celery's ScheduleEntry defaults last_run_at
to the same instant. remaining_estimate then asks dateutil for
rrule.after(last_run_at) (redbeat/schedules.py:93), which is exclusive —
so the occurrence sitting exactly on dtstart is passed over and the first run
is delayed by a full interval.

For rrule('MINUTELY', interval=3, count=4) created at 21:19:58, due_at is
21:22:58 rather than 21:19:58:

AssertionError: datetime.datetime(2017, 12, 9, 21, 22, 58, tzinfo=datetime.timezone.utc)
             != datetime.datetime(2017, 12, 9, 21, 19, 58, tzinfo=datetime.timezone.utc)

This is the first of the two problems @concreted separated out in
#74 (comment).

What's in here

tests/test_issue_74.py holds two tests:

• test_first_occurrence_is_due_at_dtstart — the bug, marked
  @unittest.expectedFailure so CI stays green. unittest reports an
  unexpected success as a failure, so the day this is fixed the suite says so.
• test_first_occurrence_is_delayed_by_one_interval — pins today's behaviour, so
  a fix can't pass the first test without also having to touch this one
  deliberately.

At fix time

This is a triage artifact, not a permanent home. Whoever fixes this should move
the first test into tests/test_schedules.py, drop the expectedFailure
marker, rename it for the behaviour it checks rather than the issue number,
delete test_first_occurrence_is_delayed_by_one_interval, and delete this file.
The same note is in the class docstring.

Not addressed here: the second problem in #74 — an rrule whose occurrences are
all in the past is never scheduled at all (due_at is None, score is
-1). That one is a design question about whether missed occurrences should
fire, not a clear defect, so I left it for the maintainer.