Investigate whether replacing `async_trait` with `dynosaur` yields perf improvements

### Is your feature request related to a problem or challenge?

I was looking through axum's changelog and I stumbled upon this benchmark, where removing `async_trait` and using Rust's 1.75's native AFIT yielded a pretty substantial performance improvement.

https://github.com/tokio-rs/axum/pull/2308#issuecomment-1880186615

### Describe the solution you'd like

I want to investigate whether removing this trait in favor of native AFIT / [dynosaur](https://crates.io/crates/dynosaur) when boxed futures are needed improves query performance. I'll test with https://github.com/alamb/datafusion-benchmarking/

### Describe alternatives you've considered

N/A

### Additional context

N/A

No DataFusion files or tests are named. Start by reviewing the linked axum benchmark and the datafusion-benchmarking project, then identify the relevant async_trait usage and benchmark the native AFIT/dynosaur alternatives. Done means documenting whether replacing the trait improves query performance.

Written by the indexing model from the issue text.