should control the maximum memory usage

### Describe the enhancement requested

`ArrowVectorIterator::consume` only concern about the `ArrowVectorIterator#targetBatchSize` but not the memory used.

Let's suppose a scenario:
1. A database table with a longtext field;
2. All of the rows in the table are larger than 500M;
3. `ArrowVectorIterator#targetBatchSize` is 256;

We need at least for 125G memory to get one record batch.

So I think it's better add a parameter such as `maxBufferSize` to control the memory usage.

Start with ArrowVectorIterator::consume and the existing targetBatchSize handling. Determine how a maxBufferSize limit should interact with record-batch construction, then define tests for large variable-length values; done means iteration respects the memory limit without requiring an oversized batch.

Written by the indexing model from the issue text.