Add regression tests for RichText serialization across all three supported shapes

[Copilot]

RichText can be a plain string, a list, or a typed subclass — but no tests verified these shapes round-tripped correctly through InputRichMessage/InputRichBlock serialization. Subtypes with extra required fields (e.g. RichTextDateTime) were silently dropping those fields.

Changes

• Merged RichText Dictionaryable support from PR Make RichText Dictionaryable #2610: adds to_dict() to all RichText subclasses and a RichText.richtext_to_dict() helper that dispatches across all three shapes; updates all InputRichBlock subclasses to use it instead of the direct .to_dict() call that broke on strings and lists.

• Added regression tests in tests/test_types.py covering:

  • test_rich_text_plain_string_serialization — plain str through InputRichBlockParagraph → InputRichMessage
  • test_rich_text_nested_list_serialization — mixed [str, RichTextBold] list
  • test_rich_text_nested_richtext_serialization — single RichTextBold object
  • test_rich_text_datetime_serialization — RichTextDateTime asserting unix_time and date_time_format are present in output

# Previously would raise AttributeError or silently drop fields
dt = types.RichTextDateTime(text='Jan 1', unix_time=1740000000, date_time_format='short')
block = types.InputRichBlockParagraph(text=dt)
d = types.InputRichMessage(blocks=[block]).to_dict()
# Now correctly produces:
# {'blocks': [{'type': 'paragraph', 'text': {'type': 'date_time', 'text': 'Jan 1',
#              'unix_time': 1740000000, 'date_time_format': 'short'}}]}