fix: list/set sampling optimization bug in _recursive_datetime_check

[Jah-yee]

Good day,

I found and fixed a bug in the _recursive_datetime_check method in DocumentBase.

Bug Description

The original code has a list/set sampling optimization that only checks the first element to decide whether to process the entire collection:

# Original buggy code
if first_checked is first_item:
    return obj

Problem: When list elements are BaseModel objects, _recursive_datetime_check modifies them in-place (via __dict__), not returning a new object. This causes first_checked is first_item to always be True, making the code incorrectly assume no conversion is needed and skipping all elements after the first one.

This was confirmed by the test file tests/test_group_profile_datetime_check.py (see test test_list_datetime_conversion):

• topics[0]: Converted ✅
• topics[1], topics[2]: Not converted ❌

Fix

1. Track converted fields: In BaseModel processing, track which datetime fields were actually converted
2. Add conversion marker: Attach _datetime_converted_fields marker to detect actual changes
3. Check marker for BaseModel: In list/set sampling logic, check the marker instead of object reference

Testing

• The existing test test_list_datetime_confirm_conversion verifies this fix
• All datetime fields in list elements should now be properly converted

感谢你们的奉献，希望能提供帮助。如果我解决得有问题或有待商妥的地方，请在下面留言，我会来处理。

Warmly,
Spark Lab Scout