[Java] reduce heap footprint of ValueVectors

In some scenarios, we hold lots of value vectors in memory eg. during join, aggregation. The heap analysis shows that the costs are as follows for a simple IntVector (used VisualVM on mac) :

IntVector : 80 bytes

vector.types.pojo.FieldType : 41 bytes

vector.types.pojo.Field : 40 bytes

IntReaderImpl : 28 bytes

I'll use this Jira to track ways to reduce the heap usage.

**Reporter**: [Pindikura Ravindra](https://issues.apache.org/jira/browse/ARROW-5482) / @pravindra
#### Related issues:
- [[Java] remove FieldReader from ValueVector](https://github.com/apache/arrow-java/issues/411) (is a parent of)
- [[Java] add ValueVector constructors that take a Field object](https://github.com/apache/arrow/issues/21932) (is a parent of)

_{**Note**: *This issue was originally created as [ARROW-5482](https://issues.apache.org/jira/browse/ARROW-5482). Please see the [migration documentation](https://github.com/apache/arrow/issues/14542) for further details.*}

先阅读 Java ValueVector 实现以及相关的“remove FieldReader from ValueVector” issue (#411)，然后使用 VisualVM 重现列出的 IntVector 堆配置文件。完成标准是：测得报告中的 ValueVector 内存开销有所降低，并记录受影响的场景和权衡。

由索引模型根据 Issue 内容生成。