[Java][Flight] Add caching of FlightClients to JDBC

### Describe the enhancement requested

When getFlightInfo reports a list of endpoints, the driver internally creates a FlightClient for each endpoint that reports a Location. These clients can benefit from being cached/pooled since a server will likely re-use endpoints for subsequent queries. This should likely be a configuration option.

The driver should also check if the Location at an endpoint is the same as the original location used when calling getFlightInfo and avoid creating a new FlightClient in this case.

Investigate using https://github.com/ben-manes/caffeine for caching.

### Component(s)

FlightRPC, Java

首先跟踪 Java JDBC 驱动的 getFlightInfo 路径，并调查它如何为返回的 Locations 创建 FlightClients。按照建议调查 Caffeine，然后定义用于客户端缓存或池化的配置选项，并验证原始 Location 会被重用，而不会创建重复的客户端。完成的标准是：后续查询会重用端点客户端，并且原始 Location 不会导致创建额外的客户端。

由索引模型根据 Issue 内容生成。