libexpr: store ExprLambda data in Expr::alloc

[Radvendii]

Motivation

For big-picture motivation, see the tracking issue

This PR moves the ExprLambda data into our allocator

• saves 32 bytes per ExprLambda which has Formals ({ foo, bar }: rather than x: )
• before:
  • 8 bytes: Formals *
  • 4 bytes: padding in ExprLambda before formals
  • 24 bytes: std::vector
  • 8 bytes: ellipsis bool + padding
• after:
  • 2 bytes: nFormals
  • 2 bytes: ellipsis bool + hasFormals bool + padding
  • 8 bytes: Formal *

And everything else stays the same.

Alternative Design

Presumably we were using a Formals * here rather than Formals directly so that we don't need to take up the extra space in the common case that there are no formals. One effect of using Formals * rather than Formals directly has been that we don't need to take up the extra space to store a Formals in the common case that there are no formals.

This PR does not make that case any worse, because we can fit the extra data (ellipses + hasFormals + nFormals) in the space that was padding before.

However, we could also choose to put all formals in one array and index into it, reducing the 8 byte Formal * to a 4 byte index.

IMO this is an optimization to look into in the future, when we start putting things in indexable vectors. (e.g. Exprs)

Context

• tracking issue

---

Add 👍 to pull requests you find important.

The Nix maintainer team uses a GitHub project board to schedule and track reviews.

[Ericson2314]

I love the way it looks now, thank you very much @Radvendii!

[xokdvium]

This should have a check in case there a more than 65k formals. We must fail gracefully.

[xokdvium]

This would overflow the heap buffer otherwise. Let's please not do this.

[xokdvium]

This can do a silent truncation from size_t -> uint16_t.