Benchmarks baseline seems to be stuck in the past

[AdamGS]

Many recent benchmarks runs I've seen use the same commit as their baseline (9b7f792, which is almost a week old), regardless of if they are rebased/updated or not.

For example, in #5026 it seems like some benchmarks use the right baseline (FineWeb NVMe) while others use this older one (TPC-H, Clickbench)

Examples:

• Arc<Box<_>> a struct array's children #4989 (comment)
• performance[vortex-datafusion]: support pushing down filters on struct columns #5024
• change gharchive query to reduce memory blowup, disable Delta #5026 (comment)

[AdamGS]

It was just a failing benchmark, @joseph-isaacs noted that it might make sense to split up some of the benchmarks so they can make progress independently.