Fix DateTimeSensor crash when target_time renders to a datetime

[1fanwang]

target_time is a template field, so it is only rendered after __init__ runs. DateTimeSensor validated and normalized it in the constructor — acting on the un-rendered Jinja expression — and poke() parsed self.target_time directly with timezone.parse(). When target_time is templated and the Dag uses render_template_as_native_obj=True, the field renders to a datetime, and poke() crashed:

TypeError: argument 'input': 'DateTime' object cannot be cast as 'str'

The deferrable path already read the value through _moment, which handles both str and datetime, so only the synchronous sensor was affected. The constructor now stores target_time verbatim, and the datetime→ISO-string normalization it used to do is relocated to poke() (the post-render execute path) rather than dropped, so self.target_time stays a string once the sensor runs. _moment still parses and validates the rendered value, and both sensors route through it. __init__ only assigns the template field now, in line with the burn-down in #70296.

related: #70296

Testing Done

Operator-level repro (real Dag, real render, real poke) — before the fix:

target_time after render: DateTime(2020, 1, 1, 0, 0, 0, tzinfo=Timezone('UTC')) DateTime
Traceback (most recent call last):
  ...
  return timezone.utcnow() > timezone.parse(self.target_time)
TypeError: argument 'input': 'DateTime' object cannot be cast as 'str'

After the fix:

target_time after render: DateTime(2020, 1, 1, 0, 0, 0, tzinfo=Timezone('UTC')) DateTime
poke -> True

test_date_time.py: 13 passed. The two normalization guards — poke() turns a native-rendered and a constructor-passed datetime into its ISO string — fail on the pre-fix source and pass after.

---

Was generative AI tooling used to co-author this PR?

• Yes (please specify the tool below)

Generated-by: GitHub Copilot CLI following the guidelines

Concerns addressed

[shahar1]

Almost there,
Also - please ensure enough test coverage for all edge cases, thanks!