Reduce Internal boilerplate

[janmasrovira]

This pr modifies the HasExpressions class to allow traversing direct subexpressions rather than only leaf expressions. I've added the lens library as a dependency to have access to generic traversals defined in Control.Lens.Plated.

[lukaszcz]

This PR switches to a different lens library and uses more general library functions to implement some traversals over program representation. I think once it's done we should benchmark it to have an idea how this affects performance.

[janmasrovira]

This PR switches to a different lens library and uses more general library functions to implement some traversals over program representation. I think once it's done we should benchmark it to have an idea how this affects performance.

afaik the implementation is equivalent so performance should be the same. It's just that lens is a bigger library.
I've run the benchmark on the stdlib and it seems to indicate that.

hyperfine --warmup 1 --prepare 'juvix clean' 'juvix typecheck index.juvix' 'juvix-main typecheck index.juvix' Benchmark 1: juvix typecheck index.juvix
  Time (mean ± σ):     436.9 ms ±   7.3 ms    [User: 874.8 ms, System: 103.0 ms]
  Range (min … max):   427.7 ms … 447.0 ms    10 runs

Benchmark 2: juvix-main typecheck index.juvix
  Time (mean ± σ):     438.6 ms ±   7.0 ms    [User: 883.1 ms, System: 88.4 ms]
  Range (min … max):   426.2 ms … 447.9 ms    10 runs

Summary
  juvix typecheck index.juvix ran
    1.00 ± 0.02 times faster than juvix-main typecheck index.juvix

I included the lens library because it has the functions that I needed and it was easier for me this way. Then we have two options, we use lens from now on, or we copy to our codebase the (relatively small) fragment that we need for this pr. That's something that we can discuss on a call next week.

[paulcadman]

I included the lens library because it has the functions that I needed and it was easier for me this way. Then we have two options, we use lens from now on, or we copy to our codebase the (relatively small) fragment that we need for this pr. That's something that we can discuss on a call next week.

There's microlens-pro that provides Prisms / Isos if that's what you need. (this is not an endorsement).

[lukaszcz]

afaik the implementation is equivalent so performance should be the same. It's just that lens is a bigger library. I've run the benchmark on the stdlib and it seems to indicate that.

It's good it doesn't matter, but such things can give you nasty surprises, like with this uniplate thing. Because someone might've implemented some accessor the wrong way, which is say 10x times slower, which makes the whole program say 2x slower because it's used all over the place.

But fortunately no nasty surprises here.