run `may_dangle` benchmarks on v2

[alejandro-vaz]

we are trying to assess what performance benefits does may_dangle have

run cargo bench and cargo bench --features may_dangle, and open a PR with no changes and the table as description

if there are barely any / no performance improvements, maybe we should just remove it instead of having arbitrary features with no ROI

if benchmarks don't really reflect the use case where this feature might perform better, add some bench for it that shows that it makes a difference, if you can find one

[alejandro-vaz]

I tried benchmarking it but at least in my machine benchmarks are wildly imprecise

I'm getting +-10% swings across recompilations