Please address the comments from this code review:

## Overall Comments
- The new `content` construction for the abstract (`"消息列表：\n```text".join([...]) + "```"`) will insert the header between every item instead of once at the beginning and also omits a newline after ```text; consider building it explicitly, e.g. `"消息列表：\n```text\n" + "".join(...) + "```"`.
- The newly added `<SUMMARY>` section in `self.train["content"]` is missing a closing `</SUMMARY>` tag even though the schema text implies a tag pair; it would be clearer and safer to wrap the summary content with both opening and closing tags.

## Individual Comments

### Comment 1
<location> `amrita/plugins/chat/handlers/chat.py:205-206` </location>
<code_context>
-                            self._dropped_messages + dropped_part, split_role=True
-                        )
-                    ]
+                content=(
+                    "消息列表：\n```text".join(
+                        [
+                            f"{it}\n"
</code_context>

<issue_to_address>
**issue (bug_risk):** Using `"消息列表：\n```text".join(...)` will repeat the header between every message instead of prefixing it once.

Because the header string is used as the `join` separator, you get `seg1 + header + seg2 + ...` rather than a single header before the full content. If you want one header and one code block wrapper, build the string once around the concatenated items, e.g.:

```python
content = (
    "消息列表：\n```text\n"
    + "".join(
        f"{it}\n" for it in text_generator(
            self._dropped_messages + dropped_part, split_role=True
        )
    )
    + "```"
)
```
</issue_to_address>

### Comment 2
<location> `amrita/plugins/chat/handlers/chat.py:477` </location>
<code_context>
                 .replace("{user_name}", str(event.sender.nickname))
             )
             + "\n</SYSTEM_PROMPT>"
+            + f"\n<SUMMARY>{data.memory_abstract if config.llm_config.enable_memory_abstract else ''}"
         )
         async with MemoryLimiter(self.data, self.train) as lim:
</code_context>

<issue_to_address>
**issue (bug_risk):** The `<SUMMARY>` tag is opened but never closed, which contradicts the schema description and may confuse downstream parsing.

Here `<SUMMARY>` is treated as a paired tag in the schema, but only the opening tag is emitted and no closing `</SUMMARY>` is added. This can break consumers expecting well-formed tags. Consider wrapping the content as a full pair:

```python
summary = data.memory_abstract if config.llm_config.enable_memory_abstract else ""
self.train["content"] = (
    ...
    + "\n</SYSTEM_PROMPT>"
    + f"\n<SUMMARY>{summary}</SUMMARY>"
)
```
</issue_to_address>