fix: preserve merged block when frame deallocation reaches the top order

[mevinagrise]

This PR fixes a FrameAllocator deallocation edge case where a fully merged block could be lost when buddy merging reached the top order. It aligns the implementation with the robust "merge-then-insert" pattern used in Heap::dealloc.

Specific changes include:

• Restricted buddy merging to valid higher orders only.
• Ensured the final merged block is always inserted back into the free list after deallocation.
• Added a regression test covering the case where two adjacent top-order blocks are freed and should remain reusable.

Why this is useful:

• Without this fix, a deallocation at the merge boundary can silently lose reusable memory.
• The new test makes the regression easy to catch in the future.
• The change is limited to FrameAllocator deallocation behavior and does not alter the public API.

Testing:

• Ran cargo test.
• Ran cargo test --doc.

[jiegec]

Please add a regression test.

[Copilot]

Pull request overview

Fixes an edge case in the buddy-style FrameAllocator deallocation logic where reaching the maximum order during buddy merging could cause the final merged block to never be reinserted into the free list (silently losing reusable memory). The change brings FrameAllocator deallocation in line with the “merge-then-insert” approach already used elsewhere (e.g., heap dealloc behavior).

Changes:

• Prevent buddy-merge attempts beyond the highest valid order.
• Always reinsert the final merged block into the appropriate free list after merging completes.
• Add a regression test that reproduces the “top-order merge loses block” scenario.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/frame.rs	Adjusts merge loop bounds and ensures the final merged block is inserted back into the free list.
src/test.rs	Adds a regression test covering deallocation/merge behavior at the maximum order boundary.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[mevinagrise]

Added a regression test in test_frame_allocator_merge_final_order in src/test.rs.

It allocates two adjacent top-order blocks, deallocates both of them, and asserts that allocating the same size succeeds again, which reproduces the lost-free-block case this PR fixes.