fix(openai): tool_call delta 重复携带 id/type，修复严格客户端兼容性

部分 OpenAI 兼容客户端要求每个 tool_call delta chunk 都包含 id 和 type
字段，否则会将后续 delta 视为无效。通过 block_to_tool_id 映射在
ContentBlockStart 时记录 tool_id，确保后续 ToolCallDelta 能正确回填。

tests/core/api_format/conversion/test_cli_conversion.py

@@ -857,15 +857,21 @@ def test_tool_call_stream_index_stable_across_deltas() -> None:
 
     # 收集所有 tool_calls chunk
     tc_indices: list[int] = []
+    tc_ids: list[str] = []
     for event in all_events:
         for choice in event.get("choices", []):
             tcs = choice.get("delta", {}).get("tool_calls")
             if isinstance(tcs, list):
                 for tc in tcs:
                     tc_indices.append(tc["index"])
+                    tc_ids.append(str(tc.get("id") or ""))
 
     assert len(tc_indices) >= 2, f"expected at least 2 tool_call chunks, got {len(tc_indices)}"
     # 同一个 tool call 的所有 chunk 必须使用相同的 index
     assert all(
         idx == tc_indices[0] for idx in tc_indices
     ), f"tool_call index should be stable, got: {tc_indices}"
+    assert all(tc_id == "fc_001" for tc_id in tc_ids), (
+        "tool_call id should be repeated on every delta for strict OpenAI-compatible clients, "
+        f"got: {tc_ids}"
+    )