allow building a "fat sysimage" with all the stdlibs in it

[KristofferC]

This is an alternative to #60139 where instead of trying to put the stdlibs into the first sysimage (which requires running their top-level code in the output-o process which doesn't run inits), we do like PackageCompiler and build a new sysimage on top of the "base" one where we can load already precompiled stdlibs. This allows us to put e.g. REPL into the sysimage which does not work the "normal" method.

❯ ./julia -e '@time using REPL; @time using Pkg'                                 
  0.002100 seconds (499 allocations: 35.359 KiB)
  0.000432 seconds (373 allocations: 30.516 KiB)

The makefile stuff is completely AI generated.

cc @pankgeorg

[topolarity]

Isn't this missing generate_precompile.jl?

Seems that it never runs on this code path

[KristofferC]

That's already run on the base image, and we build on top of that so why would we want to run that again?

[topolarity]

Isn't this using build_sysbase_$1, which is not the final base sysimage? I thought that would be $$(build_private_libdir)/sys$1.$(SHLIB_EXT)

[KristofferC]

Oh, hm, I actually wasn't aware there were two such sysimages being created. So yes, I think you are right.

[KristofferC]

Do you know the reason for not running generate_precompile.jl directly in the "base" sysimage?

[KristofferC]

This is a bit finicky, e.g.

Stdlibs total  ────────────────  5.223498 seconds
Collecting and executing precompile statements
└ Collect (Basic: Assertion failed: ((stream->type == UV_TCP || stream->type == UV_NAMED_PIPE || stream->type == UV_TTY) && "uv_write (unix) does not yet support other types of streams"), function uv__check_before_write, file src/unix/stream.c, line 1297.

[4079] signal 6: Abort trap: 6
in expression starting at /Users/kc/julia/contrib/generate_precompile.jl:10
__pthread_kill at /usr/lib/system/libsystem_kernel.dylib (unknown line)
Allocations: 5369068 (Pool: 5368927; Big: 141); GC: 8
/bin/sh: line 1:  4079 Abort trap: 6