feat(core): support variant reads via arrow-rs

[wirybeaver]

Closes #630.

Summary

Adds read compatibility for Hudi Variant columns by using arrow-rs/parquet-rs Variant support rather than adding a hudi-rs-specific Variant model.

What changes are included in this PR?

Dependencies: Arrow-rs Variant

• Enables parquet/variant_experimental in the workspace dependency.
• Uses the existing parquet::variant::{VariantType, VariantArray} APIs available in the current Arrow/Parquet 57 dependency line.

Core: Schema / Read Compatibility

• Detects Hudi Avro records with logicalType = "variant" during Avro-to-Arrow conversion.
• Represents Variant columns as Arrow Struct fields with arrow.parquet.variant extension metadata.
• Preserves the existing public read API that returns Arrow RecordBatch values.

Avro

• Validates canonical Variant child fields such as metadata, value, and optional typed_value.
• Adds tests showing Avro-decoded Variant columns can be converted with VariantArray::try_new.
• Covers child lookup by field name rather than position.

Parquet

• Relies on parquet-rs to read/write the Parquet Variant logical type through Arrow extension metadata.
• Adds tests for reading a real temporary Variant Parquet file and projecting the Variant column by top-level name.

Out of scope

• Variant write support for Hudi tables.
• New hudi-rs-specific public Variant APIs.
• SQL/query functions for nested Variant values.

Test Plan

• make format-rust
• cargo test -p hudi-core avro_to_arrow --lib
• cargo test -p hudi-core file_group::base_file::parquet::tests --lib
• cargo test -p hudi-core --lib
• make check-rust
• make test-rust
• cargo +1.91.1 check --workspace --all-targets --all-features
• git diff --check

Python checks were not run locally because ruff, mypy, and uv are not installed in this environment.