refactor: split flake.nix into multiple files

[link2xt]

There are more things to split out, cleanup and refactor e.g. around rustPlatform and rust toolchains, but i'm not going to do it now, so this PR is finished.

I tried to split out everything without changes and refactoring, the only minimal changes are for relative paths and replacing manifest.version with inherit version where the version is passed as an argument.

[Septias]

CI filters only on the Nix file and not on the subfolder nix/ so CI will not run on changes to these files. Also, the formatter would only check the format of the top-level flake file, so the folder needs to be added. I propose this fix:

--- a/.github/workflows/nix.yml
+++ b/.github/workflows/nix.yml
@@ -5,11 +5,13 @@ on:
     paths:
       - flake.nix
       - flake.lock
+      - nix/**
       - .github/workflows/nix.yml
   push:
     paths:
       - flake.nix
       - flake.lock
+      - nix/**
       - .github/workflows/nix.yml
@@ -26,7 +28,7 @@ jobs:
-      - run: nix fmt flake.nix -- --check
+      - run: nix fmt flake.nix nix/ -- --check

[Septias]

Using nix eval and Claude, I verified that the output paths are similar for 30 of 40 packages, and the 10 changed derivations are mostly due to changes to env.src which should be unproblematic. (I could not test the 6 Darwin packages that don't evaluate on my x86_64 machine). Same output paths (drvPaths) mean the build will be the same. To unblock the waiting PRs #8337 #8349 #8632, I think it is fine to merge this with only the fixed CI. If you do so, I can give the green checkmark.

There are quite a few things to be improved, though:

1. Lots of code duplication between the files in nix/
2. The instantiation in cross-rust-package, leads to ~14 full nixpkgs instantiations. Costly during evaluation.
3. Nix-filter should be used for the packages to prevent unnecessary rebuilds (due to changes on readme, etc.)
4. Nixpkgs should not point to master, since these builds are often not cached yet and could lead to stdenv builds (very costly).
5. Some inputs are (partly) discouraged (flake-utils, ~nearsk)
6. Lots of smaller changes (dead code, no default-build, nixpkgs-fmt is archived, …)

I propose splitting these into multiple follow-up PRs. I can take care of them if wanted.

[link2xt]

Some inputs are (partly) discouraged (flake-utils, ~nearsk)

I tried to get rid of naersk is the past, can be reused: #8281
But there were some problems with windows cross-compilation and i didn't want to risk windows build changing.

[Septias]

minor: your last commit has a typo btw.