Remove pkgs aliases

[nbp]

This branch aims at keeping the same hashes, but only removing aliases reported by the static analysis added in #10851 .

This branch aims at solving one of the problem which should be fixed to make the #10851 hypothesis hold, which is that all packages are defined in the first "generation" of Nixpkgs evaluation.

Thus transforming the following attribute accesses:

   +-----------+
   |           |
   |  nixpkgs  |
   |           |        +-----------+
   |   gen 3   |        |           |
   |           |        |  nixpkgs  |
   +----^------+        |           |        +-----------+
        |               |   gen 2   |        |           |   nix-build -A foo
        +---------------+           |        |  nixpkgs  |                 +
  foo_4 = callPackage…  +----^------+        |           |                 |
                             |               |   gen 1   <-----------------+
                             +---------------+           |
                         foo = foo_4;        +-----------+

Into something like:

                        +-----------+
                        |           |
                        |  nixpkgs  |
                        |           |
                        |   gen 2   |
                        |           |
                        +----^------+
                             |
                             +------------------+
                       foo_4 = callPackage…     |
                                             +-----------+
                                             |           |   nix-build -A foo
                                             |  nixpkgs  |                 +
                             +--------------->           |                 |
                             |               |   gen 1   <-----------------+
                             +---------------+           |
                         foo = self.foo_4;   +-----------+

These changes are likely to bit-rot over time, thus this pull request is a meta issue / staging area for patches which are pending for review.

cc @edolstra

update: rephrase, to make statements clearer

[vcunat]

/participate

[domenkozar]

Can we come up with a test that will check this is true for future commits?

[nbp]

@domenkozar Yes, my initial plan was to use the static analysis to ensure that the number issues does not grow. But we can only do that as soon as #10851 is landed, and as soon as the number of issues is low (no bad idioms remain).

[abbradar]

Those changes seem both trivial (and shouldn't change hashes if I understand correctly) and affecting fast-moving target (all-packages.nix). I suppose we should merge this ASAP and push further changes like this directly to master to ease burden on @nbp and others who need to keep the branch mergeable -- what do you think?