Provide four `void *` tethers per `struct splinter_slot` to enable persistent model context and vocabulary.

[splinterhq]

Each slot will get:

void *tether_1;
void *tether_2;

The library will provide:

splinter_attach_tether(const char *key, void *src_ptr, unsigned int tether_number);
splinter_detach_tether(const  char *key, unsigned int tether_number);

splinter_unset(), splinter_create(), splinter_get_slot_snapshot() and other affected artifacts will be updated. Heap allocation will be managed, but occur entirely on the client side.

This will allow Splinter to manage multiple intrinsically-safe context windows simultaneously with so many different paths for governance.

[splinterhq]

Okay. I spent some serious think time on this and I think the best way to go forward is like this:

• We implement 4 tethers, not 2. RWKV will need at least 3, 4 puts us safe for compatibility across attention architectures.
• Each one gets a corresponding size_t and pid_t initialized to zero.
• When a client attaches an allocated pointer to a tether, they update the size_t with the sizeof() the type cast to, and writes its PID to the pid_t holder. This lets the helper functions be sure of what they ultimately call free() on.
• All janitorial work is done by the two functions.
• Snapshots report the size / pid holding the tether.

So you'll be able to allocate and attach 4 things to each slot. Just don't expect changes to the pointers to also increment the slot's epoch - they're just proxies.