Add macOS support (Apple Silicon and Intel)

[aguvener]

Add macOS Support (Apple Silicon and Intel)

Draft status: NEON (ARM64) path in progress. ARM64 currently ships a scalar (optionally vDSP-accelerated) fallback, so performance trails x86_64 AVX2 for now.

Summary

Adds macOS build support on Apple Silicon (arm64) and Intel (x86_64) with automatic Homebrew dependency detection and architecture-aware flags.

Highlights

• Detects host architecture and applies AVX2/FMA only on x86_64, keeping Docker builds unchanged.
• Introduces phase/src/models/simd_compat.h to provide AVX2 vectors with a scalar fallback that preserves the existing HMM API.
• Fixes macOS linker order by placing -L search paths before -l libraries.

Testing

• Done: macOS 14 (ARM64)
• Pending: macOS Intel (x86_64)
• Pending: Linux x86_64 regression check

Technical notes

• make now gates AVX2/FMA flags behind an x86_64 host check; other architectures emit a skip message instead of failing.
• Linker flags are regrouped so search paths precede dependent libraries, resolving ld failures on macOS.
• The SIMD layer falls back to scalar math on arm64 while keeping the vector API identical to the AVX2 implementation.

Compatibility

Platform	Architecture	SIMD path	Status
Linux	x86_64	AVX2	Pending verification
macOS	x86_64	AVX2	Pending verification
macOS	ARM64	Scalar fallback	Verified (slower than AVX2)

Next steps

• Port the SIMD layer to ARM NEON before leaving draft.
• Confirm builds on macOS Intel and Linux x86_64.
• Continue enabling Linux ARM64 builds with matching SIMD support.

Reviewer checklist

• macOS Intel: build and exercise the AVX2 path.
• Linux: ensure legacy builds still succeed.

macOS setup

brew install boost htslib openssl@3 libdeflate
make

Scalar fallback keeps correctness while NEON optimisations are prepared.