ID-based safety for small objects

[nobugs-hare]

Idea: if we have a pointer to a small object, we can avoid maintain.ing those potentially very large lists of back pointers. In particular, we can:

• have allocator have a function max_object_boundaries_guaranteed_size()
• on alloc, if size <= 2^18 and size <= max_object_boundaries_guaranteed_size(), it is going to be ID-based alloc, otherwise - it is going to be list-based.
  • if it is list-based - control block stays "as is"
  • if it is ID-based - we have 8 bytes of ID before the alloc, that's it
• reserve one bit in both owning and soft ptrs to indicate it is ID-based or list-based; ID-based is used only if object_boundaries_guaranteed is returned (and if alloc size <= 2^18, see below)
• list-based logic stays pretty much "as is", except that one bit and except for moving list-based logic into non-inlinable functions
• for ID-based items we change layout of soft_ptrs completely; we have a pointer, plus 19 bits, and 64 bits for ID; out of 19, 1 bit goes for id-or-list-based flag, and we have 18 bits for offset-between-owning_ptr-and-real_ptr (and if allocation is over 2^18 in size, we enforce list-based model).
• for each conversion from soft_ptr into naked_ptr (or dereference etc.) - we compare ID field within ptr and ID in the beginning of the allocated block; if they don't match - we are accessing a dead object, it is an exception.

IMPORTANT: there are drawbacks of ID-based schema; in particular - relocatable allocs are not possible. As a result, ID-based stuff MUST be OPTIONAL (global #define)