tighten to_json adapter_name + exercise warm-up path in test

CodeRabbit follow-up on 8dd34d1:

- to_json(): switch adapter_name fallback from truthy to explicit
  `is not None`. This matches upstream's `??` semantics (preserve
  stored "" instead of rewriting it) AND avoids triggering a lazy
  `self.adapter.name` resolution when `_adapter_name` was set to ""
  but no `_adapter` is bound — that combination would raise
  RuntimeError from the adapter property.
- test_should_not_edit_placeholder_to_empty_during_llm_warmup:
  the original stream `["Hello world"]` never actually exercised
  the empty-edit guard path — by the time the edit loop fired the
  buffer already had real content. Now the stream yields a bare
  whitespace chunk, pauses long enough for the (10 ms) edit loop
  to fire, then yields the real content. A regression that emits
  empty intermediate edits would now fail here.

https://claude.ai/code/session_01XE1bMoQ5BjCvgi1iLGKKhG

Author: claude

src/chat_sdk/thread.py

@@ -744,7 +744,10 @@ def to_json(self) -> dict[str, Any]:
             "channelVisibility": self._channel_visibility,
             "currentMessage": self._current_message.to_json() if self._current_message else None,
             "isDM": self._is_dm,
-            "adapterName": self._adapter_name if self._adapter_name else self.adapter.name,
+            # Explicit `is not None` matches upstream's `??` behavior (preserve
+            # "" if that's what was stored) and avoids triggering lazy adapter
+            # resolution when `_adapter_name` was set but no `_adapter` is bound.
+            "adapterName": self._adapter_name if self._adapter_name is not None else self.adapter.name,
         }
 
     @classmethod

tests/test_thread_faithful.py

@@ -449,9 +449,18 @@ async def test_should_not_edit_placeholder_to_empty_during_llm_warmup(self):
         adapter = create_mock_adapter()
         state = create_mock_state()
 
-        thread = _make_thread(adapter, state)
-        text_stream = _create_text_stream(["Hello world"])
-        await thread.post(text_stream)
+        # Simulate the warm-up path: a whitespace-only chunk arrives before the
+        # real content, and the background edit loop fires fast enough to see
+        # it. Without the warm-up chunk the test never exercises the empty-edit
+        # guard — it'd just post "Hello world" immediately.
+        thread = _make_thread(adapter, state, streaming_update_interval_ms=10)
+
+        async def _stream() -> AsyncIterator[str]:
+            yield " "
+            await asyncio.sleep(0.05)
+            yield "Hello world\n"
+
+        await thread.post(_stream())
 
         markdown_edits = [content for _, _, content in adapter._edit_calls if isinstance(content, PostableMarkdown)]
         assert markdown_edits, "expected at least one PostableMarkdown edit"