[C++] Run-end encoded data validation could be stricter

### Describe the enhancement requested

The `ValidateRunEndEncodedChildren` function uses ordered comparisons in two places to validate the run-end encoded data:

1. To check there are at least as many values as run ends: https://github.com/apache/arrow/blob/8b2336058c1dd5eba3293ab736cfbe8e0c38dc2b/cpp/src/arrow/util/ree_util.cc#L199-L202
2. To check that the last run-end does not overpass the logical offset and length: https://github.com/apache/arrow/blob/8b2336058c1dd5eba3293ab736cfbe8e0c38dc2b/cpp/src/arrow/util/ree_util.cc#L218-L224

It seems that the current checks can let through some programming errors. An example is https://github.com/apache/arrow/issues/47029 where the JSON C++ reader would read the integration data as having logical length 7 even though the generated run-ends were much larger.

Is there a reason for not doing equality testing for these checks?

### Component(s)

C++, Integration