fix(databricks-langchain): don't repeat streamed Responses text

[MarcusKoolaard]

Fixes #464.

Summary

_convert_responses_api_chunk_to_lc_chunk suppressed the terminal full-text block only when the immediately previous chunk was response.output_text.delta. Providers close a message with response.output_text.done and response.content_part.done first, so the check never fired and the completed text was appended on top of the streamed deltas — every reply came back doubled.

Those two terminators now count as "already streamed" when the item_id matches.

Test plan

• New parametrized unit test covering both terminators before output_item.done; existing tests only covered a delta immediately preceding it, which is why this passed CI.

• pytest tests/unit_tests/test_chat_models.py — 108 passed, 2 skipped.

• Offline repro of the documented event order through _stream with a stubbed client:

  before: [{'text': 'hello'}, {'text': ' world'}, {'text': 'hello world', 'annotations': []}] -> 'hello worldhello world'
  after : [{'text': 'hello'}, {'text': ' world'}]                                             -> 'hello world'
  

• Verified live against Unity AI Gateway (Azure Foundry GPT-5.6 and system.ai.claude-haiku-4-5) through a LangGraph agent with tool calls: streamed deltas and the final message now agree.