Initial stab at JuliaC trimmability

[cmhamel]

Reworking how init and load time hooks are handled to allow for JuliaC trim safety.

• An environment flag called JULIA_MPI_TRIM_SAFE is introduced to branch in parts of the package that aren't trimmable.
• Each file that had exisiting add_load_time_hook! and add_init_hook! methods have a new method called _setup_load_time_<file_name> rather than relying on the module level arrays of hooks which caused juliac errors under trim.

With these changes I was able to compile the following program into a 2.7Mb binary that works!

using MPI

function @main(ARGS)
    MPI.Init()
    comm = MPI.COMM_WORLD
    rank = MPI.Comm_rank(comm)
    size = MPI.Comm_size(comm)
    print(Core.stdout, "Hello world, I am rank $(MPI.Comm_rank(comm)) of $(MPI.Comm_size(comm))\n")

    N = 5
    root = 0

    if MPI.Comm_rank(comm) == root
        print(Core.stdout, " Running on $(MPI.Comm_size(comm)) processes\n")
    end
    MPI.Barrier(comm)

    if MPI.Comm_rank(comm) == root
        A = [i*(1.0 + im*2.0) for i = 1:N]
    else
        A = Array{ComplexF64}(undef, N)
    end

    MPI.Bcast!(A, root, comm)

    dst = mod(rank+1, size)
    src = mod(rank-1, size)

    N = 4

    send_mesg = Array{Float64}(undef, N)
    recv_mesg = Array{Float64}(undef, N)

    fill!(send_mesg, Float64(rank))

    rreq = MPI.Irecv!(recv_mesg, comm; source=src, tag=src+32)

    sreq = MPI.Isend(send_mesg, comm; dest=dst, tag=rank+32)

    stats = MPI.Waitall([rreq, sreq])

    MPI.Finalize()
    return 0
end

Here is the set of JuliaC commands I used

using JuliaC
build_path = joinpath(@__DIR__, "build")
src_path = joinpath(@__DIR__)
@show build_path
@show src_path
rm(build_path; force=true, recursive=true)

img = ImageRecipe(
    output_type    = "--output-exe",
    file           = "$src_path/src/MyApp.jl",
    trim_mode      = "safe",
    add_ccallables = false,
    verbose        = false,
)

link = LinkRecipe(
    image_recipe = img,
    outname      = "$build_path/my_app"
)

bun = BundleRecipe(
    link_recipe = link,
    output_dir  = build_path # or `nothing` to skip bundling
)

compile_products(img)
link_products(link)
bundle_products(bun)

which I built with

JULIA_MPI_TRIM_SAFE=true julia +1.12 --project=@. make.jl

I have not tried to test every function within the package. I view this as an initial push towards allowing for MPI aware small binaries.

The test suite passes locally as well as the test suite for PartitionedArrays

This addresses issue #931.