refactor: aabb + p64 rewired — ALL 372 raw intrinsics eliminated aabb (69→0): F32x16 operators + simd_min/max/le/ge, scalar SSE fallback p64 (18→0): scalar array AND/XOR/popcount (LLVM auto-vectorizes, zero deps) 372/372 intrinsics eliminated across 12 HPC files + p64. Only 1 remaining: _mm_prefetch in jitson_cranelift (JIT hint, not data path). All 1510 ndarray tests + 23 p64 tests pass. https://claude.ai/code/session_01ChLvBfpJS8dQhHxRD4pYNp#80