fix(types): document multichannel as valid speaker-separation for conversational summary model

[pylrn]

Problem

Closes #71 (cross-repo — originally filed against assemblyai-node-sdk but the same root cause applies here).

SummarizationModel.conversational requires at least one speaker-separation mode to be enabled. The docstring listed only speaker_labels and dual_channel, leaving out multichannel — the recommended, non-deprecated alternative.

This caused confusion: users who followed the deprecation guidance and switched to multichannel=True saw what appeared to be an undocumented or unsupported configuration, even though the API accepts it correctly.

Changes

assemblyai/types.py

• Updated the SummarizationModel.conversational docstring to include multichannel in the "Required Parameters" list.
• Added a note that multichannel is the recommended option and dual_channel is deprecated — consistent with the deprecation note already present elsewhere in the SDK.

tests/unit/test_summarization.py

• Added test_conversational_summary_with_speaker_separation, a parametrized regression test covering all three valid speaker-separation flags:

  • multichannel=True (the previously undocumented case)
  • speaker_labels=True
  • dual_channel=True

  Each variant verifies that no client-side ValueError is raised and that the correct fields are forwarded to the API.

Testing

All 9 tests in test_summarization.py pass (6 pre-existing + 3 new):

tests/unit/test_summarization.py::test_conversational_summary_with_speaker_separation[multichannel] PASSED
tests/unit/test_summarization.py::test_conversational_summary_with_speaker_separation[speaker_labels] PASSED
tests/unit/test_summarization.py::test_conversational_summary_with_speaker_separation[dual_channel] PASSED
======================== 9 passed in 0.20s ========================