add %eval_nix% primop

[Radvendii]

add an %eval_nix% primop which calls out to the nix C++ bindings. At the moment, this goes through JSON and therefore requires the nix to fully evaluate

Partly addresses #1431

[vkleen]

I wonder if we can make the flake outputs more configurable to avoid the heavy dependency footprint in e.g. nix run github:tweag/nickel if one is just looking for a minimal Nickel interpreter. But that's probably better tackled in a separate PR.

[yannham]

Nice! The footprint isn't as bad as I thought (it doesn't add to much nix-experimental-related code).

[yannham]

@Radvendii is there anything left to do to merge this PR (beside rebasing, now there's a conflict)? I fear this drifts further away from master

[Radvendii]

There's a phantom bug which only occurs in CI, Not even on my local x86_64-darwin machine which is using the same version of nix to build , and building the same derivation.

I tried to track down the phantom bug using valgrind, but the nix package for valgrind is broken, and when I used an old version it didn't actually work. It seems from internet searches that valgrind is just kind of broken on mac os?

Given that this is only a problem with a single test, I've filed the bug with upstream and disabled tests for now.