Improve confidence in sandboxing: Towards LBM-images

[svenssonjoel]

Currently arbitrary C pointers can live happily inside the LBM runtime system either as part of a Custom Type or as a byte-array shared with LBM from the C side through an extension.

I propose we outlaw all arbitrary C pointers in the LBM runtime system. This would increase my confidence in the SANDBOXING that LBM offers, however maybe naively.

There are a few positive effects of this outlawing:

1. Certain things are now a double indirection, ptr->heap_cell->lbm_memory. These could turn into a direct ref.
   Arrays are currently like this: ptr-to-array -> [arbitrary-c-pointer | LBM_TYPE_ARRAY]
   |
   position in LBM memory or address created in C.

This is a big change and as such it could lead to increased debugging burden for a while after its implementation.

[svenssonjoel]

The (1.) positive effect mentioned cannot be implemented as the defragmem and the explicit free operations require that the intermediate cell exists and is unique (cannot be duplicated).