Convert to sparse array by mapping over components

[jishnub]

Structured matrix types may be converted to sparse matrices efficiently, so it makes sense to map sparse to the components of a StructArray

On master

julia> using LinearAlgebra, SparseArrays, StructArrays, BenchmarkTools

julia> D = Diagonal(Float64[1:4000;]);

julia> S = StructArray{ComplexF64}((D, D));

julia> @btime sparse($S);
  47.009 ms (16 allocations: 192.64 KiB)

This PR

julia> @btime sparse($S);
  74.021 μs (18 allocations: 407.02 KiB)

I have structured this as an extension module to make it easier to move SparseArrays to an extension in the future. I've made this an extension.

[jishnub]

Test failures are as seen in #274, and seem to be unrelated to this PR

Edit: tests appear to pass after the latest commit. The failure on nightly is due to changes in how NamedTuples are displayed in the docstrings, and is unrelated to this PR.

[piever]

Thanks for the PR and sorry for the long wait! I've rebase this in #289 (we just a small change to use createinstance to generate the sparse array. If that makes sense to you, I can go ahead and close this one and merge #289.

[jishnub]

Yes, makes sense, thanks!

[piever]

Perfect, this was merged in #289