free arrays

[gasche]

Right now, arrays are never freed. I first thought it was reasonable (heap locations allocated by mutable variables where never freed either, and the (reasonable) idea was to imagine that we had a GC), but if we start having heap-heavy tests like linked lists it would make sense to start thinking of freeing heap-allocated arrays.

We could have a free x instruction that would deallocate the array memory, but then other people that would have the array value at hand would fail on access. There is no really nice semantics there. I would be inclined to just have accesses/assignments to freed arrays get stuck; we could also have a valid(x) operator that returns true if the array still exists, and try to speculatively remove these tests.