(Potential speedup) Avoid cloning snapshots

[fkrause98]

Once we have everything set up to run benchmarks, we should test if avoid cloning snapshots every time we do a far_call improves performance.
An initial idea is to use a CoW pointer to only clone the snapshot when we need to mutate it.

[Oppen]

The CoW is unlikely to help in all but the static calls. Most programs will end up writing something, so they will end up cloning. I think we'll need something else. That said, nicely spotted, since the flamegraphs confirm cloning snapshots are a bottleneck.