Investigate call-by-need semantics for `force`-`delay`

Currently the following UPLC pseudocode:

```
let y = delay e
in f (\_ -> force y) + g (\_ -> force y)
```

will evaluate `e` twice if both `f` and `g` evaluate `force y` once.

If we used some sort of call-by-need semantics specifically for `delay` and `force`, `e` would only be evaluated twice.

This is worth investigating, I believe.