Better error for outside references

[infinisil]

Motivated by @teto in #64 (comment), improves the error messages when a path outside a package directory is referenced:

- pkgs/by-name/aa/aa: File package.nix at line 1 contains the path expression "../." which may point outside the directory of that package.
  This is undesirable because it creates dependencies between internal paths, making it harder to reorganise Nixpkgs in the future.
  Alternatives include:
  - If you are creating a new version of a package with a common file between versions, consider following the recommendation in https://github.com/NixOS/nixpkgs/tree/master/pkgs/by-name#recommendation-for-new-packages-with-multiple-versions.
  - If the path being referenced could be considered a stable interface with multiple uses, consider exposing it via a `pkgs` attribute, then taking it as a attribute argument in package.nix.
  - If the path being referenced is internal and has multiple uses, consider passing the file as an explicit `callPackage` argument in `pkgs/top-level/all-packages.nix`.
  - If the path being referenced is internal and will need to be modified independently of the original, consider copying it into the pkgs/by-name/aa/aa directory.

[teto]

The This is undesirable because it creates dependencies between internal paths, making it harder to reorganise Nixpkgs in the future. part is nice to have. As for the rest, it depends how you prefer to manage the project but one solution is to do like shellcheck and reference a wiki page like "https://www.shellcheck.net/wiki/SC2155" which can be edited without needing another release.

[infinisil]

Yeah I'd really like that better, maybe let's keep it simpler for now though

[philiptaron]

I kind of hate the - prefix, but I suppose consistency is better.

[infinisil]

@philiptaron I think it's kind of important for multiple failures: https://github.com/NixOS/nixpkgs-check-by-name/blob/bed457ebdd725a032d248fc74cb91366372eaa21/tests/multiple-failures/expected

[infinisil]

The after release workflow didn't fire automatically yet again, just like before

[infinisil]

I think this is the culprit: https://github.com/orgs/community/discussions/54574#discussioncomment-10119733