The comment for the line removed by this patch states that "A threaded
perl build needs glibc/libpthread_nonshared.a, which is not included
in bootstrapTools".  This is not the reason why it is necessary --
libpthread_nonshared.a part of the bootstrap-files on many platforms.
The reason why this workaround was needed is because nuke-refs erases
the rpath from libpthread.so, and unpack-bootstrap-tools.sh neglects
to undo this using patchelf.  Let's apply patchelf to libpthread.so,
so we aren't leaving dangling /nix/store/eeee.... refs in
libpthread.so, and remove the enableThreading=false override since it
is no longer necessary.

Additional background:

The enableThreading=false attribute not only removes the
"-Dusethreads" from the invocation of perl's Configure script, it also
patches the script like this:

  # We need to do this because the bootstrap doesn't have a static libpthread
  sed -i 's,\(libswanted.*\)pthread,\1,g' Configure

Although this prevents the perl interpreter from linking against
libpthread.so, it does not prevent the libraries bundled with the
interpreter from doing so.  Time/HiRes/HiRes.so will still try to link
against libpthread.so, and is only prevented from doing so by the fact
that unpack-bootstrap-tools.sh neglects to patchelf out all of the
nuke-refs'd rpaths.  When all of the nuke-refs'd paths are patchelf'ed
out, HiRes.so links against libpthread.so in the bootstrap tools, and
the stdenv bootstrap fails:

  Can't load '/nix/store/7ny8kmppjzx6kx4xr5kphjz8fqqrlsll-perl-5.34.1/lib/perl5/5.34.1/x86_64-linux/auto/Time/HiRes/HiRes.so' for module Time::HiRes: libpthread.so.0: cannot open shared object file: No such file or directory at /nix/store/7ny8kmppjzx6kx4xr5kphjz8fqqrlsll-perl-5.34.1/lib/perl5/5.34.1/XSLoader.pm line 93.
  at /nix/store/7ny8kmppjzx6kx4xr5kphjz8fqqrlsll-perl-5.34.1/lib/perl5/5.34.1/x86_64-linux/Time/HiRes.pm line 94.
  Compilation failed in require at lib/Autom4te/FileUtils.pm line 42.
  BEGIN failed--compilation aborted at lib/Autom4te/FileUtils.pm line 42.
  Compilation failed in require at bin/autom4te line 46.
  BEGIN failed--compilation aborted at bin/autom4te line 46.
  make[1]: *** [Makefile:2259: tests/wrapper.in] Error 2
  make[1]: *** Waiting for unfinished jobs....

Perl implements its own custom library loading routines for its
modules, which likely do not fully implement rpath searching.

This problem was revealed during the course of writing PR #161925,
which depends on this PR in order to pass CI.

… else

The current unpack-bootstrap-tools.sh wields patchelf selectively,
guided by unexplained criteria.  Specifically, it applies

patchelf --set-interpreter to:

  $out/bin/* $out/libexec/gcc/*/*/* (except */liblto*)

and patchelf --set-interpreter --set-rpath to:

  $out/lib/librt-*.so $out/lib/libpcre*

The provenance of these lists is unclear.  Some libraries which never
had an rpath to begin with (like librt.so) are given an rpath by this
script, while other libraries (like libbz2.so) are left with dangling
/nix/store/eeee...'s.

This makes it very difficult to troubleshoot the behavior of the
bootstrap-tools unpacker when porting to new platforms, where it is
likely the case that patchelf is being debugged concurrently with
nixpkgs.

This commit changes unpack-bootstrap-tools.sh to use patchelf *only*
to replace /nix/store/eeee... references created by nuke-refs, and to
replace *all* such references except where an exception is clearly
described and explained.  Currently there are two such exceptions:

  1. libgcc_s.so is exempted because its rpath leaks into the final
     stdenv, so we cannot patchelf its rpath without creating a
     fordbidden stdenv-final->bootstrap-tools requisite.

  2. libpthread.so and libc.so are exempted from --set-interpreter
     because the ELF interpreter of these files does not matter (they
     are libraries, not executable programs, and unlike ld.so they are
     not meant to play both roles), and because an unresolved bug in
     patchelf causes corruption if it is used to set the interpreter
     here:

     NixOS/patchelf#368

Making sure to patchelf away all of the nuke-refs'd paths will make
the bootstrapping process more robust and less fragile in general.  In
particular, it will prevent situations like #169746 where the only
thing preventing a bug from manifesting was the fact that the
arbitrary choice of which libraries to patchelf had been made in a
particular way.  Situations like that make things more fragile,
because now the particular choice which used to be arbitrary has
become mandatory, and neither that requirement nor the reason for it
is documented anywhere.

This commit requires #169746 (which is included in this PR) in order
to pass CI.

…f#380)