Asan fix

[tupek2]

The original bug was a teardown use-after-free.

A State object keeps a raw pointer to its DataStore. Normally that is fine because the DataStore is still alive when the State destructor runs. But ASan found a case where a copied State outlived the last owner of the DataStore. When that late-destroyed State ran its destructor, it called back into DataStore::try_to_free() through a dangling pointer. That is the crash.

The first fix added a lifetime token so a State could tell whether its DataStore was already gone. That part was correct. But the first version also changed the destructor path so it stopped calling try_to_free() in normal cases. That avoided the crash, but it also broke one of gretl's important behaviors: freeing state memory once a step is inactive and no external handles are keeping it alive.

The final fix keeps the lifetime token, but narrows the logic:

• when a State is destroyed or reassigned, it first saves the old datastore pointer, old step, and old lifetime token
• then it clears or rebinds the State
• then it only calls try_to_free() if the saved token says the old DataStore is still alive

This gives both required behaviors:

• if the DataStore is already destroyed, the State destructor does nothing dangerous
• if the DataStore is still alive, normal try_to_free() logic still runs, so gretl can evict inactive state data from memory

The added tests check both sides:

• the ASan regression test proves we no longer touch a dead DataStore
• the new eviction tests prove state payloads are still actually dropped when the last external handle goes away

[chapman39]

why not store dataStore_itself as a weak_ptr? which eliminates the use of a raw pointer and may solve this lifetime issue as well.

[chapman39]

i tested this and can confirm it eliminates the heap use after free issue