core/lmproto: allow null logprobs and usage on stream chunks

Providers send logprobs=null (and sometimes usage=null) on SSE choices;
accept None so decode does not fail mid-stream.

src/plyngent/lmproto/openai_compatible/model.py

@@ -245,7 +245,8 @@ class DeltaMessage(Struct):
 class ChunkChoice(Struct):
     index: int
     delta: DeltaMessage
-    logprobs: dict[str, Any] | Unset = UNSET
+    # APIs often send logprobs=null on stream choices.
+    logprobs: dict[str, Any] | None | Unset = UNSET
     finish_reason: FinishReason | None | Unset = UNSET
 
 
@@ -255,4 +256,5 @@ class ChatCompletionChunk(Struct):
     created: int
     model: str
     choices: list[ChunkChoice]
-    usage: dict[str, Any] | Unset = UNSET
+    # Final usage chunk may omit or null usage depending on provider.
+    usage: dict[str, Any] | None | Unset = UNSET

tests/test_lmproto/test_sse_parse.py

@@ -47,6 +47,17 @@ def test_delta_accepts_null_content_and_reasoning() -> None:
     assert delta.reasoning_content is None
 
 
+def test_chunk_choice_accepts_null_logprobs() -> None:
+    payload = (
+        b'{"id":"1","object":"chat.completion.chunk","created":0,"model":"m",'
+        b'"choices":[{"index":0,"delta":{"content":"x"},"logprobs":null,'
+        b'"finish_reason":null}],"usage":null}'
+    )
+    chunk = msgspec_decode_chunk(payload)
+    assert chunk.choices[0].logprobs is None
+    assert chunk.usage is None
+
+
 def msgspec_decode_chunk(payload: bytes) -> ChatCompletionChunk:
     import msgspec