Remove ARPACK and derived functions from julia repo

[andreasnoack]

Instead, it will become a proper package, see https://github.com/JuliaLinearAlgebra/Arpack.jl, and will use binaries from https://github.com/JuliaLinearAlgebra/ArpackBuilder. This will simplify the build process of Julia a little bit.

When moving the ARPACK related functions from base to stdlib, we decided to name the package IterativeEigensolver to avoid using the name of the underlying library. Hence, it might seem to contradict that decision that the name of the new package is Arpack. However, the idea is to create a SparseLinearAlgebra umbrella package that would be the right entry point for most users. The Arpack function would mainly be a repo for hosting the actual wrapper functions.

[ViralBShah]

Also, we have not actually used the name IterativeEigenSolvers in any released version of Julia. So, there isn't much code out there depending on it.

[staticfloat]

Great work guys. This has been a long time coming. Shoutout to @andreasnoack, @ViralBShah and @ararslan for chipping away at bits and pieces of this over the years. I am looking forward to a fortran-free Julia one of these days.

[ararslan]

Bravo! 👏 Thanks so much for taking this on, Andreas.

[ararslan]

Looks like you'll need

--- a/test/precompile.jl
+++ b/test/precompile.jl
@@ -224,7 +224,7 @@ try
                      Base.PkgId(m) => Base.module_build_id(m)
                  end for s in
                 [:Base64, :CRC32c, :Dates, :DelimitedFiles, :Distributed, :FileWatching, :Markdown,
-                 :Future, :IterativeEigensolvers, :Libdl, :LinearAlgebra, :Logging, :Mmap, :Printf,
+                 :Future, :Libdl, :LinearAlgebra, :Logging, :Mmap, :Printf,
                  :Profile, :Random, :Serialization, :SharedArrays, :SparseArrays, :SuiteSparse, :Test,
                  :Unicode, :REPL, :InteractiveUtils, :OldPkg, :Pkg, :LibGit2, :SHA, :UUIDs, :Sockets]))
         @test discard_module.(deps) == deps1

[andreasnoack]

It looks like the tests are currently failings because Documenter depends on Compat and Compat depends on IterativeEigensolvers. I've opened JuliaLang/Compat.jl#578 to resolve this.

[KristofferC]

You can just add that branch of Compat to the docs project if you want to test the doc building right away.

[andreasnoack]

You can just add that branch of Compat to the docs project if you want to test the doc building right away.

Fancy new package manager. I've tried that so let's see how it goes.

[ararslan]

The macOS failure is #26725 and the Circle build looks like it got OOM killed.

[ViralBShah]

Seems good to merge.

[KristofferC]

The Compat PR should likely be merged and a new version tagged first. Otherwise there will be no Compat compatible with the latest build.