Fix HeapMeta alignment on 32-bit platforms using C11 union

[mstorm]

Summary

This PR addresses alignment issues for HeapMeta on 32-bit architectures by restructuring it using a C11 anonymous struct within a union. This enforces a 16-byte size without altering its alignof requirement, preventing potential alignment-related crashes on 32-bit systems (where malloc typically returns 8-byte aligned addresses).

Fixes #342

Type of Change

• Bug fix (non-breaking change which fixes an issue)

Test Plan

• Ran make lib to verify that the static assertion sizeof(HeapMeta) == 16 passes across different configurations.
• Ran make test (1619 tests passed) to ensure no regressions in allocator logic and overall functionality.
• Verified memory layout and assembly code generation (ARM64, -O2) to confirm zero runtime overhead.

Test Configuration

• Compiler: Apple clang version 17.0.0
• Build type: Default (Release)
• Platform(s): macOS (Apple Silicon / arm64), logic verified for ILP32 (32-bit arm/x86) compatibility.

---

Technical Analysis (For Reviewers)

We compared three approaches: (1) Manual #ifdef padding, (2) ZL_ALIGNED(16), and (3) C11 union.

• Efficiency: Assembly analysis shows that the union approach generates identical machine code to the original struct. Field access (stp/ldp) and pointer arithmetic remain optimal.
• Safety: Unlike ZL_ALIGNED(16), which raises alignof(HeapMeta) to 16 and could cause runtime assertion failures with 8-byte aligned malloc pointers, the union approach keeps the original alignment requirement (8) while guaranteeing the 16-byte size.
• Portability: Uses standard C11 features, avoiding non-standard attributes or platform-specific macros.

Metric	Original Struct	C11 Union
sizeof	16 (64-bit) / 8 (32-bit)	16 (Both)
alignof	8	8
Codegen	Baseline	Identical