Extend GPUArrays compat to 12

[lkdvos]

Extends the compat for GPUArrays to 11, 12 and for JLArrays to 0.3, 0.4.

With GPUArrays 11 / JLArrays 0.3, the tensor network testset in test/allocator.jl errors with "output tensor must not be aliased with input tensor". Once the buffer is warm, temporaries handed out by JLBufferAllocator are reported as aliased by Base.mightalias. With GPUArrays v12.1.0 and JLArrays v0.4.2, test/allocator.jl (205 tests) and the JLArray part of test/gpu.jl (89 tests) pass locally.

This only resolves once Strided and StridedViews allow these versions too: QuantumKitHub/Strided.jl#93 and QuantumKitHub/StridedViews.jl#59. Even then, CI keeps resolving to GPUArrays 11 for now, because CUDACore and AMDGPU in the test environment still cap it at 11.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
see 35 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.