feat: AllocateAligned — per-allocation GPU-aligned sub-allocation

[kolkov]

Summary

• AllocateAligned(size, alignment uint32) for GPU-aligned sub-allocation
• TLSF-style over-allocate: O(1), zero heap allocs, zero overhead when alignment <= 1
• Supports all GPU alignment requirements: Vulkan (256), DX12 (4-65536), Metal (64-256)
• SyncAllocator.AllocateAligned thread-safe wrapper

Details

Per-allocation alignment is a prerequisite for GPU integration (ADR-050 Phase 3). Every real GPU allocator (gpu-allocator, gpu-alloc, TLSF) uses per-allocation alignment. Upstream C++ OffsetAllocator and Rust offset-allocator don't have it — we're the first port to add it.

Implementation over-allocates by up to alignment - 1 bytes, then adjusts the returned offset. Free() works unchanged (uses Metadata node index, not Offset). Waste is <6% for the dominant 256-byte alignment case.

Test plan

• 12 new tests: various alignments (4, 32, 64, 256, 512, 65536), multiple aligned allocations with overlap check, fast paths (alignment 0/1), alignment > size, non-power-of-2 panic, exhaustion, coalescing, sync, concurrent
• 2 new benchmarks: AllocateAligned256 (~50ns/op), AllocateAligned1 (fast path)
• FuzzAllocAligned: 2.5M+ executions, 0 failures, 97 interesting inputs
• All 35 tests pass, 0 lint issues, 0 heap allocations
• Cross-platform build: Windows, Linux, macOS

[codecov]

Welcome to Codecov 🎉

Once you merge this PR into your default branch, you're all set! Codecov will compare coverage reports and display results in all future pull requests.

ℹ️ You can also turn on project coverage checks and project coverage reporting on Pull Request comment

Thanks for integrating Codecov - We've got you covered ☂️