Fix indexing for chains in different threads

[penelopeysm]

Closes #153

If there were 6 threads and 9 chains, the previous code (

AbstractMCMC.jl/src/sample.jl

Lines 441 to 445 in 5a3b155

	chainidxs = if i == nchunks
	((i - 1) * chunksize + 1):nchains
	else
	((i - 1) * chunksize + 1):(i * chunksize)
	end

) would generate the following chainidxs for each thread

1:2
3:4
5:6
7:8
9:10
11:10

Somewhere later on (probably in the fifth thread with chainidxs = 9:10) this causes indexing errors as we only have a vector of size 9.

This PR changes it such that for the same inputs (6 threads and 9 chains) it generates the following chainidxs

1:2
3:4
5:6
7:7
8:8
9:9

The error in #153 is fixed

julia> samples = [sample_chains(chain_count) for chain_count in 1:30]
30-element Vector{Chains{Float64, AxisArrays.AxisArray{Float64, 3, Array{Float64, 3}, Tuple{AxisArrays.Axis{:iter, StepRange{Int64, Int64}}, AxisArrays.Axis{:var, Vector{Symbol}}, AxisArrays.Axis{:chain, UnitRange{Int64}}}}, Missing, @NamedTuple{parameters::Vector{Symbol}, internals::Vector{Symbol}}}}:
 MCMC chain (1000×14×1 Array{Float64, 3})
 MCMC chain (1000×14×2 Array{Float64, 3})
 MCMC chain (1000×14×3 Array{Float64, 3})
 MCMC chain (1000×14×4 Array{Float64, 3})
 MCMC chain (1000×14×5 Array{Float64, 3})
 MCMC chain (1000×14×6 Array{Float64, 3})
 MCMC chain (1000×14×7 Array{Float64, 3})
 MCMC chain (1000×14×8 Array{Float64, 3})
 MCMC chain (1000×14×9 Array{Float64, 3})
 MCMC chain (1000×14×10 Array{Float64, 3})
 MCMC chain (1000×14×11 Array{Float64, 3})
 MCMC chain (1000×14×12 Array{Float64, 3})
 MCMC chain (1000×14×13 Array{Float64, 3})
 MCMC chain (1000×14×14 Array{Float64, 3})
 MCMC chain (1000×14×15 Array{Float64, 3})
 MCMC chain (1000×14×16 Array{Float64, 3})
 MCMC chain (1000×14×17 Array{Float64, 3})
 MCMC chain (1000×14×18 Array{Float64, 3})
 MCMC chain (1000×14×19 Array{Float64, 3})
 MCMC chain (1000×14×20 Array{Float64, 3})
 MCMC chain (1000×14×21 Array{Float64, 3})
 MCMC chain (1000×14×22 Array{Float64, 3})
 MCMC chain (1000×14×23 Array{Float64, 3})
 MCMC chain (1000×14×24 Array{Float64, 3})
 MCMC chain (1000×14×25 Array{Float64, 3})
 MCMC chain (1000×14×26 Array{Float64, 3})
 MCMC chain (1000×14×27 Array{Float64, 3})
 MCMC chain (1000×14×28 Array{Float64, 3})
 MCMC chain (1000×14×29 Array{Float64, 3})
 MCMC chain (1000×14×30 Array{Float64, 3})

julia> println("Thread count: $(Threads.nthreads())")
Thread count: 6

julia> for i in eachindex(samples)
           isnothing(samples[i]) && println("BoundsError with $i chains")
       end

[mhauru]

I'm happy with the code, just wondering about whether this could be tested somehow. End-to-end is hard, but might it make sense to put the chainidxs logic in a function that could be tested with given values for nchains and number of threads?

[penelopeysm]

@mhauru the logic itself feels too basic to be worth testing to me 😅. I checked again locally with the most recent commit and it still behaves correctly, so it has been end-to-end tested, albeit not in CI.

[mhauru]

I get paranoid about off-by-one mistakes with all index arithmetic, and testing can also guard against botched future changes. If the person who wrote the previous version would have added tests we wouldn't be here. But I do see that this is very minimal, and also it's not introducing untested functionality, it's rather fixing untested functionality, so I'm okay with merging as-is. Thanks for fixing!