94c64466

Take the mutation children's locale from a Haskell derivation

An alternative to #151, same fix, taken from the Haskell builder rather than
restated. Pick one; I will close the other.

## The difference

#151 writes the values out:

```nix
localeEnv = {
  LANG = "en_US.UTF-8";
} // pkgs.lib.optionalAttrs (pkgs.stdenv.hostPlatform.libc == "glibc") {
  LOCALE_ARCHIVE = "${pkgs.glibcLocales}/lib/locale/locale-archive";
};
```

This takes them off a derivation the builder made:

```nix
localeEnv = builtins.intersectAttrs { LANG = null; LOCALE_ARCHIVE = null; } driver;
```

`driver` is `haskellPackages.sydtest-mutation-driver`, already in scope, and a
Haskell derivation exposes both as attributes. `intersectAttrs` rather than
`inherit` because the builder sets `LOCALE_ARCHIVE` only on glibc, so this
stays total elsewhere.

Nothing is restated, so it follows whatever nixpkgs decides the Haskell locale
is instead of drifting from it. It also fixes a bug in #151's version: I wrote
`hostPlatform.libc` and `pkgs.glibcLocales` where generic-builder.nix uses
`buildPlatform.libc` and `buildPackages.glibcLocales` — identical natively,
wrong when cross-compiling.

## Why not call the Haskell builder itself

That was the first thing I tried. `haskellPackages.mkDerivation` takes a closed
argument list of Cabal-package arguments and has no `...`, so it rejects
`dontUnpack` and anything else stdenv-shaped:

```
error: function at generic-builder.nix:89 called with unexpected argument 'dontUnpack'
       Did you mean postUnpack?
```

These two derivations build no Cabal package. They run test executables the
builder has already produced, so there is nothing to configure, build or
install, and routing them through it would mean suppressing nearly every phase
while still pulling GHC and the Setup machinery in. Borrowing the environment
is the part that is actually wanted.

## Same outcome, provably

On x86_64-linux the resulting derivation is **bit-identical** to #151's:

```
#151:        /nix/store/pzcn1avpzyrq6f4mq5m4abqvfjllny4c-mutation-leader-data-mutation-check.drv
this branch: /nix/store/pzcn1avpzyrq6f4mq5m4abqvfjllny4c-mutation-leader-data-mutation-check.drv
```

So all of #151's evidence carries over unchanged, and `nix build` of a real
check against this branch is a store hit that succeeds. The built derivation
carries `LANG=en_US.UTF-8` and a `LOCALE_ARCHIVE` equal to the one an ordinary
Haskell test derivation gets.

Background and the failure this fixes: NorfairKing/nix-ci#762.