Fixes the macOS ARM C++ backend build while keeping NEON SIMD enabled.

• Adds Pixie-local typed NEON load/store shims so Nim C++ emits uint8/uint32 pointers instead of void pointers for Apple arm_neon.h macros.
• Routes existing NEON load/store call sites through those shims.
• Adds the nim cpp test runner on this stacked branch, leaving the image-suite PR focused on decoder coverage.

Local checks:

• nim check --cpu:arm64 --os:macosx tests/tests.nim
• nim cpp -r tests/tests.nim