Missing embedding_metadata in response - matching engine - Vector Store

I am using the preview feature of vertexai vector store that support `embedding_metadata`.
When using the curl command I can retrieve the metadata associated with a datapoint.

However, this doesn't work with the python SDK.

The `find_neighbors` method of `MatchingEngineIndexEndpoint` accepts the parameter `return_full_datapoint` which then correctly adds the param to the api endpoint.
The issue is that when the response is cast into the class `MatchNeighbor`, which does not contain a field for `embedding_metadata`, this information is lost even though it was retrieved.

Solution is to correctly cast it and/or update the definition of `MatchNeighbor`.

The solution I'm currently using in my own fork is to add

```python
if (index_datapoint.embedding_metadata is not None):
self.embedding_metadata = index_datapoint.embedding_metadata
```

in line 280 of matching_engine.matching_engine_index_endpoint.py

从 matching_engine.matching_engine_index_endpoint.py 中的 MatchingEngineIndexEndpoint.find_neighbors 和 MatchNeighbor 响应转换开始。检查 return_full_datapoint 响应的映射方式，然后验证当 API 提供 embedding_metadata 时，它是否会保留在返回的 MatchNeighbor 中。

由索引模型根据 Issue 内容生成。