Adding support for shared_memory allocator for multiprocessing.

How feasible is it modifying `static MapObject * map_alloc(void)` (and the `...dealloc` functions) to use a custom allocator, if buffer parameter is passed to constructor, so a multiprocessing.shared_memory.SharedMemory().buf can be used, similarly to Numpy?

Start with the static map_alloc and dealloc functions named in the issue, then trace the constructor path and its buffer handling. Compare the proposed SharedMemory.buf use with the existing Numpy approach, including the C extension boundary. Done should be a clearly scoped design and feasibility result with any required tests identified.

Written by the indexing model from the issue text.