Fix GPU CI: qualify validate_pure_leaping_inputs in KernelAbstractions ext

[ChrisRackauckas-Claude]

Summary

• Master GPU CI is red with UndefVarError: validate_pure_leaping_inputs not defined in JumpProcessesKernelAbstractionsExt (last red run: SciML/JumpProcesses.jl actions run 25110242250).
• validate_pure_leaping_inputs is defined in src/simple_regular_solve.jl but not exported from JumpProcesses. The extension was calling it bare, which is invalid since using JumpProcesses only brings exported names into scope.
• Smallest possible fix: qualify the call with JumpProcesses.validate_pure_leaping_inputs(...). Matches the existing pattern in JumpProcessesOrdinaryDiffEqCoreExt.jl, which uses JumpProcesses.__jump_init.

Notes

• Please ignore until reviewed by @ChrisRackauckas.
• CPU Run Tests was already green; this only affects the GPU Tests workflow.
• GPU CI cannot be reproduced locally; the fix is verified by reading the failure log and matching it against the known-unexported function.

Test plan

• GPU Tests workflow passes on this PR.
• CPU Run Tests still green (unchanged code path on CPU).

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

Update on GPU CI status

Initial CI run completed. Result: Run Tests and CPU jobs are green (or in-progress, expected to be green); GPU Tests is red but with a different, pre-existing failure than the one this PR set out to fix.

What this PR fixed

Verified working — the test now reaches kernel compilation, which means JumpProcesses.validate_pure_leaping_inputs(...) resolves correctly. The original UndefVarError is gone.

Newly-surfaced issue (separate from this PR's scope)

Run 25629456752 shows:

LoadError: InvalidIRError: compiling MethodInstance for
  JumpProcessesKernelAbstractionsExt.gpu_simple_tau_leaping_kernel(...)
  resulted in invalid LLVM IR
Reason: unsupported call to an unknown function (call to jl_f_throw_methoderror)
Stacktrace:
 [1] rand          @ Random/src/generation.jl:114    # Julia 1.12.6
 [2] rand          @ Random/src/Random.jl:255
 [3] randexp       @ CUDACore/src/device/random.jl:339   (device_override)
 [4] count_rand    @ PoissonRandom/MLLFD/src/PoissonRandom.jl:18
 [5] pois_rand     @ PoissonRandom/MLLFD/src/PoissonRandom.jl:137
 [6] kernel body   @ ext/JumpProcessesKernelAbstractionsExt.jl:135
 ...

Line 135 of the kernel is counts[k] = pois_rand(PoissonRandom.PassthroughRNG(), rate_cache[k]).

Dispatch chain on the GPU (Julia 1.12.6, CUDACore gtlJx, PoissonRandom MLLFD):

• randexp(::PassthroughRNG) → bare randexp() (defined in PoissonRandom)
• randexp() → randexp(default_rng()) → randexp(::Philox2x32) (CUDACore overrides default_rng on device)
• randexp(::AbstractRNG) device override at CUDACore/src/device/random.jl:339 calls Random.rand(rng, Random.UInt52Raw())
• rand(rng, X) at Random.jl:255 → Sampler(rng, X, Val(1)) then rand(rng, sampler)
• rand(r::AbstractRNG, ::SamplerTrivial{UInt52Raw{UInt64}}) at generation.jl:114 → _rand52(r, rng_native_52(r))
• Somewhere on this path, Julia 1.12 has introduced a method-error path that GPUCompiler can't prove unreachable.

This is a compatibility bug between PoissonRandom's PassthroughRNG, Julia 1.12's Random stdlib, and CUDACore's device random overrides — not introduced by this PR. It would have blocked master GPU CI just the same once the validate_pure_leaping_inputs issue was past, since the failure is in kernel codegen.

Per the small-PR philosophy in CLAUDE.md, I'm keeping this PR focused on the original UndefVarError fix and tracking the new IR issue separately. The PR successfully makes incremental progress and is independently reviewable/mergeable.

🤖 Generated with Claude Code