Optimize import loading time

Please consider optimizing the imports that this library performs. On my system a single `import vertexai` takes about 1 second.

The top 3 modules that take the most time can be computed with `-X importtime`
```
$ python -X importtime -m vertexai &> x
$ sort -k 3 -n x | tail -n 3
import time: 7300 | 7300 | google.cloud.aiplatform_v1beta1.types.evaluation_service
import time: 53394 | 53394 | google.cloud.aiplatform_v1beta1.services.pipeline_service.transports.rest_asyncio
import time: 244048 | 259512 | google.api_core
```
Where the numbers are measured in microseconds (us).

Version: google-cloud-aiplatform v1.128.0

首先运行报告中的 `python -X importtime -m vertexai` 测量并对结果进行排序。检查点名的 `google.cloud.aiplatform_v1beta1` 和 `google.api_core` 导入，以确定加载路径；完成的标准是减少单次 `import vertexai` 所需的时间。

由索引模型根据 Issue 内容生成。