[Variant] Benchmark for shred_variant kernels

[alamb]

This is a draft because it needs additional shredding support (for Utf8 and timestamp, for example)

Which issue does this PR close?

• Follow on to [Variant] Define new shred_variant function #8366
• Part of [EPIC] [Parquet] Implement Variant type support in Parquet #6736

Rationale for this change

I would like to make sure variant_shred is super fast and the first part of that is benchmarking it

What changes are included in this PR?

Add a benchmark

Run it like this

cargo bench --bench variant_kernels -- shred

Are these changes tested?

I ran it manually and it is tested by clippy in by CI

Are there any user-facing changes?

No