BroadcastStyle for RectDiagonal

[jishnub]

This hooks into some of the functionality for structured matrices in LinearAlgebra, and assuages some of the concerns raised in #265. After this, sparsity is often preserved when broadcasting over a RectDiagonal:

julia> E = Eye(2,4)
2×4 Eye{Float64}

julia> 2E
2×4 FillArrays.RectDiagonal{Float64, Vector{Float64}, Tuple{Base.OneTo{Int64}, Base.OneTo{Int64}}}:
 2.0   ⋅    ⋅    ⋅ 
  ⋅   2.0   ⋅    ⋅ 

julia> E .+ 1
2×4 Matrix{Float64}:
 2.0  1.0  1.0  1.0
 1.0  2.0  1.0  1.0

[codecov]

Codecov Report

Merging #275 (08b33e1) into master (6122f5a) will increase coverage by 95.53%.
The diff coverage is 100.00%.

@@             Coverage Diff             @@
##           master     #275       +/-   ##
===========================================
+ Coverage    0.00%   95.53%   +95.53%     
===========================================
  Files           5        5               
  Lines         849      852        +3     
===========================================
+ Hits            0      814      +814     
+ Misses        849       38      -811     

Impacted Files	Coverage Δ
src/FillArrays.jl	96.51% <100.00%> (+96.51%)	⬆️

... and 4 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[dlfivefifty]

Perhaps this warrants v1.1 as its a "new feature"?

[jishnub]

Yes, this may go in the upcoming 1.4 release. The version is already bumped on master, but it hasn't been tagged yet