feat: implement bitmap-managed arena for improved cache locality

[ByteMaster2003]

Closes #22

Summary

This PR introduces arena3, a bitmap-managed memory allocator designed to replace the linked-list approach in arena2. By managing memory as a grid of 64-byte cells and tracking them via a bitmask, we improve allocation speed and significantly reduce metadata overhead.

Key Changes

• Bitmap-based Tracking: Replaced the next pointer logic with a Vec<u64> bitmap.
• Consolidated GcBox: Removed ArenaHeapItem. The GcBox<T> is now the top-level structure stored in the arena, saving 8 bytes per object.
• Optimized Allocation: Implemented bitwise scanning using trailing_zeros() for O(1) allocation on the fast path.
• Alignment Guarantees: Enforced 64-byte cell alignment to ensure predictable pointer arithmetic (Address=Base+(Index×64)).

Architectural Improvements

1. Reduced Fragmentation: By using fixed-size cells, we can quickly identify gaps without traversing the entire heap.
2. Cache Efficiency: During the "Sweep" phase, the collector only needs to scan the bitmap vector (likely residing in L1 cache) before touching the actual object memory.
3. API Parity: Maintained functional parity with arena2 signatures to allow for a drop-in replacement during the benchmarking phase.

Current Status

• Basic BitmapArena implementation.
• Fast-path allocation logic.
• Consolidated header layout.
• Sweep logic integration (pending architectural feedback).
• Arena recycling/swapping strategy.