Gibbs test slowdown

[penelopeysm]

Since #2328 was merged, the time it takes to run mcmc/gibbs is far longer than the time it used to take to run mcmc/gibbs plus experimental/gibbs.

Need to investigate the cause for this.

[mhauru]

Could you point me to a CI run where this is evident? I'm looking at this one as an example for the pre-merge times, and the times for experimental seem equally long (and wildly variable): https://github.com/TuringLang/Turing.jl/actions/runs/12325358448/job/34404526010

[mhauru]

There's definitely something weird going on between Julia v1.10 and v1.11 though, the latter being 2-3 times slower in cases I'm looking at. @yebai, you mentioned that you think this is due to a Libtask issue. Is the root of the problem known? Would it still be helpful to narrow this down, to confirm that if we leave out all the particle Gibbs tests then v1.10 and v1.11 become comparable in runtime?