implement per-chain progress bars with ProgressMeter.jl

[penelopeysm]

This PR enables per-chain progress bars when sampling multiple chains with MCMCThreads().

To do so, it switches the logging functionality to use ProgressMeter.jl which appears to be much more actively maintained.

As a bonus, it also means that when sampling with isdone (i.e., we don't know how many iterations will happen until isdone returns true) we can make use of ProgressMeter.ProgressUnknown which is a more sensible interface than the typical progress bar.

I had first tried to implement this using the existing logging setup (with ProgressLogging.jl -- see https://github.com/TuringLang/AbstractMCMC.jl/tree/py/parallel), but found myself blocked by JuliaLogging/ProgressLogging.jl#41 (I have asked on Slack to see if somebody would merge it).

Here is a video:

Screen.Recording.2025-06-27.at.01.20.51.mov

The code I ran in the video is:

using AbstractMCMC
struct M <: AbstractMCMC.AbstractModel end
struct S <: AbstractMCMC.AbstractSampler end
function AbstractMCMC.step(rng, ::M, ::S, state=nothing; kwargs...)
    sleep(0.002)
    rand(rng), nothing
end

sample(M(), S(), 500)

isd(rng, model, sampler, samples, state, i; kwargs...) = i > 500
sample(M(), S(), isd)

sample(M(), S(), MCMCThreads(), 500, 6)

TODO

• MCMCSerial
• MCMCDistributed
• Fix tests
• Is there a way to fix the per-chain progress bars not being cleared fully? keep=true doesn't seem to have the intended behaviour
• Check platform generality (I'm mildly concerned that, for example, ProgressMeter might not work on Pluto)

Closes #24
Closes #82
Would close TuringLang/Turing.jl#2264 as well

[github-actions]

AbstractMCMC.jl documentation for PR #165 is available at:
https://TuringLang.github.io/AbstractMCMC.jl/previews/PR165/