Benchmark restructuring

[AdamGS]

As of this writing, we're stuck in dependency hell where we can't upgrade our DataFusion/Arrow dependencies because we pull lance to benchmark against which is still on older version because of their dependencies.
Our benchmarks also suffer from long build times because every iteration is built (at least) with DataFusion and DuckDB which are both pretty big, and iteration times are not great.
I propose the following structure:

1. bench-vortex becomes vortex-bench and will only have the minimal shared code required for the benchmarks - like the JSON formatting, data generation, query and workload enumeration, potentially some shared CLI arguments, and ideally Vortex specific code behind a feature.
2. Create a new benches directory that has a bunch of dedicated benchmarking crates, one for each big engine. This way, we can have a different DF version that Lance.
3. Create a shared runner CLI (potentially just an xtask) that can invoke those dedicated runners, so the overall experience stays streamlined as possible.

Some of the non-sql benchmarks right now depend on post-processing to get ratios between vortex/parquet/lance, but we can shift those to the website.

[gatesn]

SGTM

[joseph-isaacs]

It would be great to keep ratios between datafusion and duckdb locally

[AdamGS]

People are welcome to follow along here