[java] cannot read datasets with many columns

### Describe the bug, including details regarding any error messages, version, and platform.

Dataset API gets stuck if one tries to read many columns (100k+) from a parquet or arrow file. The API works if the amount of columns is narrowed via projection.

**Reproduction:**

Use the code from the [example](https://arrow.apache.org/cookbook/java/dataset.html#constructing-datasets) and try to read a file with 100k columns. No batch will be loaded even if the batch size is 1.

version: arrow-dataset 11.0

### Component(s)

Java

Start with the Java Dataset API and the linked dataset construction example, reproducing the hang with a parquet or Arrow file containing 100k+ columns and a batch size of 1. Trace the read path while comparing it with a narrowed projection; done means batches load without getting stuck for the reported wide dataset case.

Written by the indexing model from the issue text.