[AutoBuild] Split log files into a separate tarball

[giordano]

Long term plan is still to do #778, but that's a lot of work, and we have little time at the moment. Simply splitting the logs into a separate tarball, instead, is a simpler short-term solution which has some very important bonuses:

• the git tree hash of the main tarball should be reproducible (as long as you use the same toolchain)
• which means we can start testing reproducibility of builds
• rebuilding a package without changing its content won't create a new artifact: less strain on the Package Storage Server.

I still need to do tests about deployment on Yggdrasil though (Edit: see JuliaPackaging/Yggdrasil#5526).

[giordano]

Uhm, the failure

[ Info: Building for x86_64-linux-gnu-libgfortran5-cxx11-libstdcxx29-julia_version+1.7.2, x86_64-unknown-freebsd
JLLs - building: Error During Test at /data/staticfloat/bb_azp/rootfs/agent_1/_work/1/s/test/jll.jl:41
  Got exception outside of a @test
  Unable to locate LibraryProduct(["libfoo"], :libfoo, String[], false, Symbol[]) within /data/staticfloat/bb_azp/rootfs/agent_1/tmp/jl_26NC7I for x86_64-linux-gnu-libgfortran5-cxx11-libstdcxx29-julia_version+1.7.2
  Stacktrace:
    [1] error(s::String)
      @ Base ./error.jl:33
    [2] (::BinaryBuilder.var"#45#49"{Bool, Vector{Product}, String, Vector{String}, Dict{Any, Any}, Platform, String, String})(dest_prefix::String)
      @ BinaryBuilder /data/staticfloat/bb_azp/rootfs/agent_1/_work/1/s/src/AutoBuild.jl:1174
    [3] mktempdir(fn::BinaryBuilder.var"#45#49"{Bool, Vector{Product}, String, Vector{String}, Dict{Any, Any}, Platform, String, String}, parent::String; prefix::String)
      @ Base.Filesystem ./file.jl:750
    [4] mktempdir (repeats 2 times)
      @ ./file.jl:748 [inlined]
    [5] rebuild_jll_package(name::String, build_version::VersionNumber, sources::Vector{DirectorySource}, platforms::Vector{Platform}, products::Vector{Product}, dependencies::Vector{Dependency}, download_dir::String, upload_prefix::String; code_dir::String, verbose::Bool, from_scratch::Bool, kwargs::Base.Pairs{Symbol, Any, NTuple{4, Symbol}, NamedTuple{(:lazy_artifacts, :julia_compat, :init_block, :augment_platform_block), Tuple{Bool, String, String, String}}})
      @ BinaryBuilder /data/staticfloat/bb_azp/rootfs/agent_1/_work/1/s/src/AutoBuild.jl:1161
    [6] rebuild_jll_package(obj::Dict{String, Any}; download_dir::String, upload_prefix::String, build_version::Nothing, gh_org::String, verbose::Bool, from_scratch::Bool)
      @ BinaryBuilder /data/staticfloat/bb_azp/rootfs/agent_1/_work/1/s/src/AutoBuild.jl:1103
    [7] (::var"#134#138"{String, String, Platform})()
      @ Main /data/staticfloat/bb_azp/rootfs/agent_1/_work/1/s/test/jll.jl:160
    [8] withenv(::var"#134#138"{String, String, Platform}, ::Pair{String, String}, ::Vararg{Pair{String, String}})
      @ Base ./env.jl:172
    [9] (::var"#133#137")(build_path::String)
      @ Main /data/staticfloat/bb_azp/rootfs/agent_1/_work/1/s/test/jll.jl:103
   [10] mktempdir(fn::var"#133#137", parent::String; prefix::String)
      @ Base.Filesystem ./file.jl:750
   [11] mktempdir(fn::Function, parent::String) (repeats 2 times)
      @ Base.Filesystem ./file.jl:748
   [12] macro expansion
      @ /data/staticfloat/bb_azp/rootfs/agent_1/_work/1/s/test/jll.jl:42 [inlined]
   [13] macro expansion
      @ /data/staticfloat/bb_azp/rootfs/usr/local/julia-1.7/share/julia/stdlib/v1.7/Test/src/Test.jl:1283 [inlined]
   [14] top-level scope
      @ /data/staticfloat/bb_azp/rootfs/agent_1/_work/1/s/test/jll.jl:42
   [15] include(fname::String)
      @ Base.MainInclude ./client.jl:451
   [16] top-level scope
      @ /data/staticfloat/bb_azp/rootfs/agent_1/_work/1/s/test/runtests.jl:72
   [17] include(fname::String)
      @ Base.MainInclude ./client.jl:451
   [18] top-level scope
      @ none:6
   [19] eval
      @ ./boot.jl:373 [inlined]
   [20] exec_options(opts::Base.JLOptions)
      @ Base ./client.jl:268
   [21] _start()
      @ Base ./client.jl:495

looks unrelated to the changes to this package in this PR, but maybe some other upstream changes in BBB broke it (but we haven't seen any widespread breakage). I'll have to dig into this.

Edit: I did a first round of investigation and I found it's precisely the change to autobuild in this PR which breaks the test. It's weird, I can't quickly tell how this is related to that test, will have to dig more.