[Java] Unexpected Allocator states created after allocating buffer whose AllocationManager has different size from the requested size

If a buffer with altered AllocationManager size (which means, the size is different from request size) is created from a Allocator, then allocator still adds the request size rather than the altered size to its accountant. As a result inconsistency will be made once the buffer is getting destroyed because at that time the actual buffer size is used for releasing.

The predefined AllocationManager implementations don't make such inconsistencies since they always grant chunks within the request size. However we have [documented](https://github.com/apache/arrow/blob/e5f3e04b4b80c9b9c53f1f0f71f39d9f8308dced/java/memory/memory-core/src/main/java/org/apache/arrow/memory/AllocationManager.java#L184-L191) that the actual size can be altered from request size in custom implementations of AllocationManager. So this kind of customization should be supported.

**Reporter**: [Hongze Zhang](https://issues.apache.org/jira/browse/ARROW-14034) / @zhztheplayer
**Assignee**: [Hongze Zhang](https://issues.apache.org/jira/browse/ARROW-14034) / @zhztheplayer
#### PRs and other links:
- [GitHub Pull Request apache/arrow#11182](https://github.com/apache/arrow/pull/11182)

_{**Note**: *This issue was originally created as [ARROW-14034](https://issues.apache.org/jira/browse/ARROW-14034). Please see the [migration documentation](https://github.com/apache/arrow/issues/14542) for further details.*}

Start with the AllocationManager size documentation linked in the issue, then trace the allocator's accounting when a buffer is created and destroyed. Review PR apache/arrow#11182 for the work already underway. Done means custom AllocationManager sizes no longer leave allocator accounting inconsistent across allocation and release.

Written by the indexing model from the issue text.