reduce built-in MemBuilders sizes

[tower120]

All built-in MemBuilder implementations can store Layout in compacted form.

[tower120]

Implemented in https://github.com/tower120/any_vec/tree/CompactLayout branch for HeapMem. Possibly introduce some overhead on into_raw_parts/from_raw_parts due to packing/unpacking Layout. Changing MemRawParts trait API probably would solve this. Or maybe overhead is not observable, benchmark needed.