Forward structure-preserving broadcasting to diag for Diagonal

[jishnub]

For example, after this PR,

julia> D = Diagonal(1:4)
4×4 Diagonal{Int64, UnitRange{Int64}}:
 1  ⋅  ⋅  ⋅
 ⋅  2  ⋅  ⋅
 ⋅  ⋅  3  ⋅
 ⋅  ⋅  ⋅  4

julia> D .* 2
4×4 Diagonal{Int64, StepRangeLen{Int64, Int64, Int64, Int64}}:
 2  ⋅  ⋅  ⋅
 ⋅  4  ⋅  ⋅
 ⋅  ⋅  6  ⋅
 ⋅  ⋅  ⋅  8

julia> using SparseArrays

julia> D = Diagonal(spzeros(2))
2×2 Diagonal{Float64, SparseVector{Float64, Int64}}:
 0.0   ⋅ 
  ⋅   0.0

julia> D .* 2
2×2 Diagonal{Float64, SparseVector{Float64, Int64}}:
 0.0   ⋅ 
  ⋅   0.0

julia> using FillArrays

julia> D = Diagonal(Fill(3, 2))
2×2 Diagonal{Int64, Fill{Int64, 1, Tuple{Base.OneTo{Int64}}}}:
 3  ⋅
 ⋅  3

julia> D .* 2
2×2 Diagonal{Int64, Fill{Int64, 1, Tuple{Base.OneTo{Int64}}}}:
 6  ⋅
 ⋅  6

I've not handled the other banded matrix types in this PR, but Diagonal is probably used far more commonly in packages anyway.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.89%. Comparing base (98723df) to head (3e4e9ca).
⚠️ Report is 2 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1423      +/-   ##
==========================================
- Coverage   93.89%   93.89%   -0.01%     
==========================================
  Files          34       34              
  Lines       15920    15928       +8     
==========================================
+ Hits        14948    14955       +7     
- Misses        972      973       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[dlfivefifty]

Is there a reason to use map instead of preprocess_broadcasted.(T, bc.args)?

(I think in old versions of Julia making "enclosures" like this was bad for performance but perhaps that's ancient history.)

[jishnub]

map on small tuples unrolls the loop. I don't think the anonymous function would be an issue here (perhaps the issue has been resolved). The call seems type-stable and well-inferred.