BlockSparseArrays.jl release to-do list

[mtfishman]

To-do list for splitting off NDTensors.BlockSparseArrays as a separate registered package BlockSparseArrays.jl:

• Remove the AnyAbstractBlockSparseArray type union in favor of an @derive macro, similar to Moshi.@derive, the Rust derive attribute for implementing traits, and ArrayLayouts.@layoutmatrix and related macros in ArrayLayouts.jl. This would basically automatically define getindex, map!, etc. as blocksparse_getindex, blocksparse_map!, etc. on a specified type or wrapper.
• Assess which NDTensors.jl sub-modules BlockSparseArrays depends on and either remove those dependencies or assess what we need to do to split off those libraries into packages as well. For example:
  • SparseArraysBase is a major dependency so we will have to release that first, see `SparseArraysBase.jl` release to-do list SparseArraysBase.jl#1.
  • BroadcastMapConversion, which converts broadcast calls to map calls (which is heavily inspired by the broadcasting code logic in Strided.jl). That library is also used in other sub-modules of NDTensors.jl, such as BlockSparseArrays and NamedDimsArrays.
  • GradedAxes is being used for things like dual and to provide some generic block axis slicing functionality that works for both graded and non-graded unit ranges.
  • TypeParameterAccessors for generically accessing type parameters, in particular it uses functionality for generically getting the type of the parent of a wrapper type. We've been planning to split that off for a while, though I think there are still some type instability issues and interface questions to decide on so I'm not sure how comfortable I am doing that right now.
  • NestedPermutedDimsArrays will be used as the output of blocks(::PermutedDimsArray).
  • There are internal "package extensions" that depend on GradedAxes and TensorAlgebra for compatibility with those libraries.
• Fix open issues in #1336.
• Reassess the minimal interface of BlockSparseArrays, particularly in light of any changes we decide to make to SparseArraysBase, which are being discussed in `SparseArraysBase.jl` release to-do list SparseArraysBase.jl#1.

[mtfishman]

I forgot about this discussion, this is pretty outdated and I think mostly implemented so I'll close it, we can open a new discussion if needed.