TEST ONLY: verify SciML/PoissonRandom.jl#72 unblocks GPU tau-leaping

[ChrisRackauckas-Claude]

This PR exists solely to run GPU CI against the proposed fix in SciML/PoissonRandom.jl#72 before that PR is released. Do not merge.

Background

Issue #588: GPU kernel compile fails with InvalidIRError: jl_f_throw_methoderror on Julia 1.12 / latest CUDACore.

Root cause is in PoissonRandom, not in JumpProcesses. GPUCompiler installs CUDACore's overrides into a Julia overlay method table via Base.Experimental.@consistent_overlay. Julia 1.12's OverlayMethodTable.findall returns overlay matches without consulting the base method table whenever the overlay fully covers the signature (Compiler/src/methodtable.jl:73–92). So @device_override Random.randexp(rng::AbstractRNG) shadows PoissonRandom's specific Random.randexp(rng::PassthroughRNG) on the device, and the override body runs with rng::PassthroughRNG. Its Random.rand(rng, UInt52Raw()) then reaches rng_native_52(::PassthroughRNG) which has no method — hence the static throw(MethodError) that GPUCompiler rejects.

SciML/PoissonRandom.jl#72 fixes this by adding the two forwarding methods PassthroughRNG needs to survive overlay shadowing:

Random.rng_native_52(::PassthroughRNG) = UInt64
Random.rand(rng::PassthroughRNG, ::Type{T}) where {T} = rand(T)

This PR

• No code change to JumpProcesses.
• Adds a [sources] override in Project.toml pinning PoissonRandom to the fix branch.
• Purpose: prove on real GPU CI that the PoissonRandom fix unblocks test/gpu/regular_jumps.jl.

Plan after CI

• If GPU Tests pass: merge Make PassthroughRNG dispatch survive overlay method-table shadowing PoissonRandom.jl#72, tag 0.4.8, close this PR without merging. JumpProcesses master will then pick up 0.4.8 automatically (existing compat is "0.4").
• If GPU Tests fail: the failure narrows the remaining problem; iterate on PoissonRandom or follow up upstream (there is a known separate CUDA.jl bug — @noinline randexp_unlikely triggers julia.get_pgcstack on Julia 1.12 — that may surface once the MethodError is unblocked).

Notes

• Please ignore until reviewed by @ChrisRackauckas.

🤖 Generated with Claude Code