Add JLArray tests for GPU-like array interface compliance

[ChrisRackauckas-Claude]

Summary

This PR adds JLArray-based tests to enhance the interface compliance testing of SimpleDiffEq.jl:

• Added JLArrays as a test dependency - JLArrays provide GPU-like arrays that catch interface violations such as improper scalar indexing or type hardcoding
• Added JLArray-based tests to interface_tests.jl - Tests verify that all OOP (out-of-place) solvers correctly support GPU-like arrays
• Added compat entries for JLArrays (0.1, 0.2, 0.3) and OrdinaryDiffEq (6)

Interface Compliance Results

All OOP solvers correctly support JLArrays:

• ✅ SimpleEuler (OOP)
• ✅ SimpleRK4 (OOP)
• ✅ SimpleTsit5 (OOP)
• ✅ SimpleATsit5 (OOP)
• ✅ GPUSimpleTsit5
• ✅ GPUSimpleATsit5

The IIP (in-place) versions use scalar indexing for performance, which is the expected design - they are CPU-optimized, while the GPU* variants are designed for GPU compatibility.

Pre-existing Interface Compliance

The package already had good interface compliance:

• ✅ BigFloat support (scalars, vectors, SVector) - tested in existing interface_tests.jl
• ✅ SVector support - tested in gpu_ode_regression.jl
• ✅ Proper use of zero(u0) and similar() for type-generic initialization
• ✅ Broadcasting in OOP versions for GPU compatibility

Test plan

• Run full test suite locally - all 23 interface tests pass (up from 15)
• CI tests pass

cc @ChrisRackauckas

🤖 Generated with Claude Code