Parquet regression: Invalid array of tuples

### Company or project name

_No response_

### Describe what's wrong

We have parquet files on s3 with a deeply nested structure:
- ~10 top level columns
- 2k columns in total
- multi-level nested structure with combination of structs/arrays

In 25.10 we could read these using `s3Cluster`, with 25.11 we get the following:

```
DB::Exception: Invalid array of tuples: tuple elements a.b and a.c.d.e have different array lengths
```

(actual column/field names omitted, but paths are similar)

The arrays indeed have different lengths, but they are completely different arrays, in a different places in the nested structure.

### Does it reproduce on the most recent release?

Yes

### How to reproduce

```
select * from s3Cluster('...', 'https://.../*.parquet')
```

### Expected behavior

_No response_

### Error message and/or stacktrace

_No response_

### Additional context

These parquet files are read successfully using all of Spark, DuckDB and Polars.