[FEATURE] Shredded Parquet Reader/Writer support for Variant type

### What is the problem the feature request solves?

Apache Spark 4.0 introduced the VARIANT semi-structured data type.
Currently, Spark writes VARIANT columns to Parquet as opaque binary blobs
(value + metadata bytes per row). This means:

- No column pruning: the entire binary blob is read even if only one sub-field is needed
- No predicate pushdown: filters on VARIANT sub-fields (e.g. payload:temp > 20.0)
cannot be pushed into the Parquet scan

Ref
https://github.com/apache/arrow-rs/issues/6736 , https://github.com/apache/parquet-format/pull/456

Although we can get details here https://github.com/apache/arrow-rs/issues/6736 , but I think DataFusion v53.0.0 does NOT depend on parquet-variant yet.

### Describe the potential solution

_No response_

### Additional context

_No response_