Fix reading variant null values in Delta Lake

[chenjian2664]

Description

In cases where a value chunk includes both null and non-null
entries, the chunk must be treated as potentially nullable.

This is the another corner case that missed in #26027

Release notes

## Delta Lake
* Fix failure when reading `null` values on `json` type columns. ({issue}`26184`)

[ebyhr]

/test-with-secrets sha=b06be75c90801490ebc04fefaa3ce32f0ab7b6ae

[github-actions]

The CI workflow run with tests that require additional secrets has been started: https://github.com/trinodb/trino/actions/runs/16253030103

[ebyhr]

Does it mean Parquet spec is wrong, or Databricks violates the spec?
https://github.com/apache/parquet-format/blob/master/VariantEncoding.md#variant-in-parquet

The value field must be annotated as required for unshredded Variant values

[chenjian2664]

Here is the meta of the test parquet file that written by Databricks:

parquet meta part-00000-3dae12c4-61bc-4177-bd36-2c936db81e90-c000.snappy.parquet

File path:  part-00000-3dae12c4-61bc-4177-bd36-2c936db81e90-c000.snappy.parquet
Created by: parquet-mr version 1.12.3-databricks-0002 (build 2484a95dbe16a0023e3eb29c201f99ff9ea771ee)
Properties:
                   org.apache.spark.version: 3.5.0
            com.databricks.spark.jobGroupId: 1752418278132_8642402946176337271_3e73399aa411479296f9bc16c62a6681
  org.apache.spark.sql.parquet.row.metadata: {"type":"struct","fields":[{"name":"id","type":"integer","nullable":true,"metadata":{}},{"name":"x","type":"variant","nullable":true,"metadata":{}}]}
             com.databricks.spark.clusterId: 1002-064054-nbosugsx
Schema:
message spark_schema {
  optional int32 id;
  optional group x {
    required binary value;
    required binary metadata;
  }
}

I think the Schema is correct

But I don't know why our NestedColumnReader#readNonNull can't read correctly cc @raunaqmorarka

[chenjian2664]

#26194