[eyalamirmusic]

Summary

• Test-only dependency: ScopedMemoryAllocations is fetched in Tests/CMakeLists.txt and linked only to the test executable, so it sits behind PROJECT_IS_TOP_LEVEL like NanoTest. Projects that add_subdirectory this repo never fetch or link it. The tests use it to check that pool and resource allocations don't touch the global heap, or touch it exactly when expected.
• New tests in Tests/Allocators/ for PMR.h and MultiPoolAllocator.h: DynamicMemory, ResourceContainer, MemoryPool, MemoryPoolResource, PMR::Object / PMR::Vector, makeShared and makeOwned.
• Fix: over-aligned allocation. MemoryPoolResource ignored the requested alignment. The alignment now goes through MemoryPool and DynamicMemory, which reserve bytes + alignment - 1 and align with std::align. The cutoff is __STDCPP_DEFAULT_NEW_ALIGNMENT__, the alignment the block storage already gets from operator new, so requests at or below it behave exactly as before.
• Fix: PMR::Object assignment. Copy and move assignment now keep the target's own resource and assign only the contents, as std::pmr containers do, so getResource() always matches the resource the wrapped object allocates from. PMR::Object also gets a real move constructor, which doesn't allocate.
• Behaviour change: smaller defaults. MemoryPoolResource used to default to 1000 pre-filled 1 MB blocks (about 1 GB at construction). It now defaults to 16 blocks of 64 KB with 1024 reserved bookkeeping slots (about 1 MB). MemoryPool's default counts changed to match (16 / 1024). Existing code that relied on the larger defaults will fall back to the heap sooner.
• New: ObjectPoolResource<T>. A pool resource with blocks of sizeof(T) that honours alignof(T), pre-filled with N blocks (default 1024).
• New: PMR::makeOwned<T>(resource, args...) returns PMR::Owned<T>, a std::unique_ptr whose deleter frees the memory back to the same resource with the size and alignment captured at creation. Converting to Owned<Base> is allowed only when Base has a virtual destructor.
• Comments: MemoryPool and the pool resources now say they are heap-backed and that pointers stay put until deallocated.

Testing

ctest: 563/563 passed locally on macOS (Apple clang). That is the existing tests plus 53 new ones.