feat: [Windows] CI for AMD RoCM support

[hiro-v]

Problem

• Lack support for AMD RoCM on Windows

Success Criteria

• CI pipeline on hosted Github action runner
• Cortex llama.cpp support for AMD RoCM

Steps BUILD

• [x] Register github action runner for Windows AMD

• [x] Install Visual Studio with C++

• [x] Install HIP SDK for Windows

• [x] choco install cmake llvm ninja strawberryperl

• [x] PATH=C:\Program Files\AMD\ROCm\5.7\bin

• [x] Build command

  cmake -S ./nitro_deps -B ./build_deps/nitro_deps
  cmake --build ./build_deps/nitro_deps --config Release
  mkdir -p build
  cd build
  cmake -G Ninja -DAMDGPU_TARGETS=gfx1100 -DLLAMA_HIPBLAS=ON -DCMAKE_C_COMPILER="C:\Program Files\AMD\ROCm\5.7\bin\clang.exe" -DCMAKE_CXX_COMPILER="C:\Program Files\AMD\ROCm\5.7\bin\clang++.exe" -DCMAKE_BUILD_TYPE=Release -DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_BLAS=ON -DBUILD_SHARED_LIBS=OFF -DWHISPER_SDL2=ON -DLLAMA_HIPBLAS=ON -DAMDGPU_TARGETS="gfx803;gfx900;gfx906:xnack-;gfx908:xnack-;gfx90a:xnack+;gfx90a:xnack-;gfx940;gfx941;gfx942;gfx1010;gfx1012;gfx1030;gfx1100;gfx1101;gfx1102" ..

• [x] Replace -lm.lib -> -lmsvcrt.lib in build/build.ninja

• [x] Run build command after the fix: cmake --build . --config Release

INFERENCE

• [x] Install HIP SDK so that it has rocblas and rocm in C:\Program Files\AMD\ROCm\5.7\bin (We can discover the ability to package this and put into the same directory similar to what we did with CUDA)
• [x] Usage

  cp build_deps/bin/zlib.dll build/
  $env:HIP_VISIBLE_DEVICES = "1" # Choose external card instead of integrated card
  $env:HSA_OVERRIDE_GFX_VERSION = "11.0.0" # AMD Radeon 7600 (however as the target is for multiple targets, it should be fine. However in order to make it work with 
  .\nitro.exe

[dan-homebrew]

Merged into https://github.com/janhq/cortex.cpp/issues/323