This saves a bit of work for the allocator at free time. I don't expect this to be a huge win but we might be able to save a small amount of CPU by doing this in a few key places:

• SharedBuffer (covers at least BSONObj and Message)
• StackAllocator (Used by StackBufBuilder when exceeding inline size)

There are a few places in query code that would benefit from this as well

• DocumentStorage::_cache (search for unique_ptr<char[]> in document.cpp)
• A handful of places in SBE use std::free or delete[]

Note that sized_free is a C23 addition and while supported by the new tcmalloc may not be available everywhere. We should consider using switching things using mongoMalloc to ::operator new(size) and ::operator delete(ptr, size) instead. Another advantage to this is that tcmalloc uses (slightly) more efficient rules for ::operator new than for malloc.