Fix bug handling Adjoints in macros

[odow]

Not sure if this is the best way to achieve this, but it seems to work.

Closes #1667.

[codecov]

Codecov Report

Merging #1698 into master will decrease coverage by <.01%.
The diff coverage is 0%.

@@            Coverage Diff             @@
##           master    #1698      +/-   ##
==========================================
- Coverage   68.52%   68.51%   -0.01%     
==========================================
  Files          30       30              
  Lines        3867     3869       +2     
==========================================
+ Hits         2650     2651       +1     
- Misses       1217     1218       +1

Impacted Files	Coverage Δ
src/parse_expr.jl	66.4% <0%> (-0.27%)	⬇️
src/macros.jl	84.18% <0%> (+0.02%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d32fce7...ee43d21. Read the comment docs.

[codecov]

Codecov Report

Merging #1698 into master will increase coverage by 0.15%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #1698      +/-   ##
==========================================
+ Coverage   68.52%   68.67%   +0.15%     
==========================================
  Files          30       30              
  Lines        3867     3943      +76     
==========================================
+ Hits         2650     2708      +58     
- Misses       1217     1235      +18

Impacted Files	Coverage Δ
src/parse_expr.jl	66.66% <100%> (ø)	⬆️
src/quad_expr.jl	70.27% <0%> (-0.49%)	⬇️
src/macros.jl	84.15% <0%> (ø)	⬆️
src/aff_expr.jl	71.92% <0%> (+0.11%)	⬆️
src/objective.jl	95.23% <0%> (+1.48%)	⬆️
src/constraints.jl	71.85% <0%> (+3.35%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d32fce7...40a2580. Read the comment docs.

[blegat]

If you replace 0.0 here:
https://github.com/JuliaOpt/JuMP.jl/blob/b78da6727823b594ef020997e6e48c3e7d15ed90/src/parse_expr.jl#L98
By Val(false), the method called is now JuMP.destructive_add_with_reorder!(::Val{false}, ::Adjoint) which calls
https://github.com/JuliaOpt/JuMP.jl/blob/b78da6727823b594ef020997e6e48c3e7d15ed90/src/parse_expr.jl#L295
The broadcast still collects the adjoint into a matrix but overloading this one may be easier.
By the way, it seems like a bug that broadcasting an adjoint lead to collect.

[odow]

I don't really know what I'm doing because I haven't played around with the broadcasting stuff, but:

julia> f(x) = x' .+ 1
f (generic function with 1 method)

julia> g(x) = Broadcast.materialize(Broadcast.broadcast(+, adjoint(x), 1))
g (generic function with 1 method)

julia> x = [1, 2]
2-element Array{Int64,1}:
1
2

julia> f(x)
1×2 Array{Int64,2}:
2  3

julia> g(x)
1×2 LinearAlgebra.Adjoint{Int64,Array{Int64,1}}:
2  3

julia> @code_lowered f(x)
CodeInfo(
1 1 ─ %1 = Base.Broadcast.materialize                                       │
  │   %2 = Base.Broadcast.broadcasted                                       │
  │   %3 = (Base.adjoint)(x)                                                │
  │   %4 = (%2)(Main.:+, %3, 1)                                             │
  │   %5 = (%1)(%4)                                                         │
  └──      return %5                                                        │
)

julia> @code_lowered g(x)
CodeInfo(
1 1 ─ %1 = (Base.getproperty)(Main.Broadcast, :materialize)                 │
  │   %2 = (Base.getproperty)(Main.Broadcast, :broadcast)                   │
  │   %3 = (Main.adjoint)(x)                                                │
  │   %4 = (%2)(Main.:+, %3, 1)                                             │
  │   %5 = (%1)(%4)                                                         │
  └──      return %5                                                        │
)

Related issues: JuliaLang/julia#25671, JuliaArrays/AxisArrays.jl#131

[blegat]

It seems to be thank to this:
https://github.com/JuliaLang/julia/blob/163b0196b12a84143388fcc0037873ef68f0beec/stdlib/LinearAlgebra/src/adjtrans.jl#L185-L186
which was introduced in:
JuliaLang/julia#25083
A special method is defined for broadcast with Adjoint vectors and scalars.
However, it seems that .+ is lowered to broadcasted followed by materialize instead of broadcast so the method is not called.
I don't understand why it's not lowered to broadcast since it is basically equivalent:
https://github.com/JuliaLang/julia/blob/163b0196b12a84143388fcc0037873ef68f0beec/base/broadcast.jl#L707