Add .github/copilot-instructions.md for agent onboarding

[Copilot]

Adds the copilot-instructions.md file to onboard the GitHub Copilot coding agent, giving it actionable context about how this repo is structured and how to work in it efficiently.

Contents

• Project overview — four fusion techniques, CPU+CUDA backends, main vs LTS-C++17 branch distinction
• Repository layout — annotated directory tree mapping key paths to their purpose
• Build system — CMake >= 3.24 requirements, configure/build commands, all key cmake -D options, CUDA arch constraints (compute_70+ enforced for CUDA < 13), Windows Ninja NVCC path workaround
• Test infrastructure — header-based test discovery mechanics (discover_tests.cmake), ONLY_CU/ONLY_CPU markers, required launch() function contract, _cpp/_cu target naming
• Code conventions — full FK_* macro table with NVCC/CPU-mode expansions, namespace fk, type aliases, gpuErrchk, clang-format settings (4-space, 120-col, LLVM-based)
• Core API patterns — executeOperations<DPPType> usage, ::build() factory pattern, key data types (Ptr2D, Tensor, Stream)
• CI matrix — self-hosted runner configs, compiler versions (GCC 13, Clang 21, MSVC 14.44/14.50, clang-cl), CUDA 12.9/13.2, env var wiring (CC/CXX/CUDACXX)
• Known issues — Windows rules.ninja NVCC path patch, clang-as-CUDA-compiler explicitly not supported despite CLANG_HOST_DEVICE macro existing

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.