Fix Apple Silicon builds: header ordering, SIMD guard, and qd toolchain

[Mohit-Lakra]

Fixes #35

This PR fixes the Apple Silicon / ARM build breakages reported in #35.

lp_solve header ordering: we now make sure the bundled lp_solve headers are picked up before any system include dirs (via src/external/lpSolve/src/Makefile and the calling src/Makevars). This avoids cases where Homebrew’s lp_types.h is found first and LPSREAL ends up missing.

PackedCSparse intrinsics: src/external/PackedCSparse/FloatArray.h only pulls in <immintrin.h> when building for AVX2 on x86, so ARM builds stay on the existing scalar path and don’t trip over x86-only headers.

qd build/toolchain consistency: the vendored qd build now compiles .cc sources with $(CXX) and the C++ flags forwarded from src/Makevars (and runs ranlib on libqd.a where needed). This avoids ABI/arch mismatches when R uses different CC/CXX settings.

After these changes, R CMD INSTALL . completes successfully on Apple Silicon (tested with Homebrew R 4.5), and there’s no behavior change expected on x86.

[Mohit-Lakra]

Hi @vissarion
When you have a moment, could you please review this and share your feedback? I’d also appreciate your perspective on whether the scope looks appropriate for merging.

Thank you.

[arnavsharma990]

hey @Mohit-Lakra Great fix for the arm64 build issue
The SIMD guard addition looks clean.

Just wondering are there any remaining AVX-specific code paths that might need guarding beyond the header include?

[Mohit-Lakra]

Hey @arnavsharma990 thanks!
At this point, I don’t think there are any other AVX-specific paths that need extra guarding. The intrinsic code was already wrapped in #ifdef __AVX2__ , so ARM never touches it. The only thing that was leaking into arm64 builds was the unconditional <immintrin.h> include.