Python: fix: ChatHistoryTruncationReducer orphans TOOL role messages (#13608)

## Summary

Fixes #12708

`ChatHistoryTruncationReducer.reduce()` can orphan `TOOL` role messages
by truncating the preceding `assistant` message that contains the
`tool_calls`, causing OpenAI to reject the history with:

```
messages with role 'tool' must be a response to a preceding message with 'tool_calls'
```

## Root Cause

`locate_safe_reduction_index()` uses
`contains_function_call_or_result()` to detect tool-related messages
during its backward scan. This function only checks `msg.items` for
`FunctionCallContent`/`FunctionResultContent` instances.

However, `TOOL` role messages can contain only text content (no
`FunctionResultContent` in `items`), which causes the backward scan to
treat them as regular messages. The truncation point then lands between
the `tool_calls` assistant message and its `TOOL` responses, orphaning
the tool results.

## Fix

Added `AuthorRole.TOOL` check to `contains_function_call_or_result()`:

```python
if msg.role == AuthorRole.TOOL:
    return True
```

This ensures **any** `TOOL` role message is recognized as part of a tool
call/result pair, regardless of whether it has `FunctionResultContent`
in its `items`.

## Test

Added
`test_locate_safe_reduction_index_tool_role_without_function_result_content`
that verifies TOOL role messages with only text content are not
separated from their tool call.

---------

Co-authored-by: giulio-leone <giulio.leone@users.noreply.github.com>
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: 4 people

python/semantic_kernel/contents/history_reducer/chat_history_reducer_utils.py

@@ -237,5 +237,12 @@ def extract_range(
 
 @experimental
 def contains_function_call_or_result(msg: ChatMessageContent) -> bool:
-    """Return True if the message has any function call or function result."""
+    """Return True if the message has any function call or function result.
+
+    Also returns True for TOOL role messages, which are always responses to
+    a preceding assistant message with tool_calls and must not be separated
+    from it.
+    """
+    if msg.role == AuthorRole.TOOL:
+        return True
     return any(isinstance(item, (FunctionCallContent, FunctionResultContent)) for item in msg.items)

python/tests/unit/contents/test_chat_history_reducer_utils.py

@@ -194,3 +194,39 @@ def test_locate_safe_reduction_index_high_offset(chat_messages_with_pairs):
     else:
         # It's fine if it returns None, meaning no valid safe reduction was found.
         pass
+
+
+def test_locate_safe_reduction_index_tool_role_without_function_result_content():
+    """Regression test: TOOL role messages without FunctionResultContent in items
+    must still be recognized as part of a tool call/result pair.
+
+    This prevents orphaning tool results when the TOOL message only contains
+    text content (no FunctionResultContent item).
+    """
+    msgs = [
+        ChatMessageContent(role=AuthorRole.USER, content="Hello"),
+    ]
+    # Assistant with tool call
+    msg_call = ChatMessageContent(role=AuthorRole.ASSISTANT, content="")
+    msg_call.items.append(FunctionCallContent(id="call1", function_name="myTool", arguments={"x": 1}))
+    msgs.append(msg_call)
+
+    # Tool result as role=TOOL but with plain text content only
+    msgs.append(ChatMessageContent(role=AuthorRole.TOOL, content="Tool result here"))
+
+    msgs.append(ChatMessageContent(role=AuthorRole.USER, content="Thanks"))
+    msgs.append(ChatMessageContent(role=AuthorRole.ASSISTANT, content="You are welcome"))
+
+    idx = locate_safe_reduction_index(msgs, target_count=3, threshold_count=0)
+    assert idx is not None
+
+    # The tool call (index 1) must be included if tool result (index 2) is included
+    kept_indices = list(range(idx, len(msgs)))
+    has_tool_role = any(msgs[i].role == AuthorRole.TOOL for i in kept_indices)
+    has_tool_call = any(any(isinstance(it, FunctionCallContent) for it in msgs[i].items) for i in kept_indices)
+
+    if has_tool_role:
+        assert has_tool_call, (
+            f"Tool result at index 2 was kept but tool call at index 1 was dropped. "
+            f"Kept indices: {kept_indices}, reduction index: {idx}"
+        )