Fix some major invalidations to improve compile times

[ChrisRackauckas]

Unitful.jl was flagged as a major invalidator of compilation downstream. Test:

# From: https://timholy.github.io/SnoopCompile.jl/stable/snoopr/
using SnoopCompile
invalidations = @snoopr begin
    using DifferentialEquations

    function lorenz(du,u,p,t)
     du[1] = 10.0(u[2]-u[1])
     du[2] = u[1]*(28.0-u[3]) - u[2]
     du[3] = u[1]*u[2] - (8/3)*u[3]
    end
    u0 = [1.0;0.0;0.0]
    tspan = (0.0,100.0)
    prob = ODEProblem(lorenz,u0,tspan)
    alg = Rodas5()
    tinf = solve(prob,alg)
end;

trees = SnoopCompile.invalidation_trees(invalidations);

@show length(SnoopCompile.uinvalidated(invalidations)) # show total invalidations

show(trees[end]) # show the most invalidated method

This method won the prize for the absolute most invalidations 🎉. But I think the bigger issue is that it simply doesn't follow Julia semantics. It fixes the types for issue #127 in a way that gives a stricter type than Julia would do in the normal cases (which is why the invalidation occurs).

After this PR, heterogeneous arrays of numbers with Quantity in there act normally, and compile times are back to normal. Here's a showcase of it being normal:

using Unitful, Test
m = u"m"
cm = u"cm"

b = Union{Complex,Float64}[0 + 0im, 0.0]
@test b + b == b
@test b .+ b == b
@test eltype(b+b) === Number

b = Number[0 + 0im, 0.0]
@test b + b == b
@test b .+ b == b
@test eltype(b+b) === Number

b = [0.0, 0.0m]
@test b + b == b
@test b .+ b == b
@test eltype(b+b) === Number

[ChrisRackauckas]

It would be good if this could get a quick review because it effects a lot of downstream compile times @sostock

[codecov-commenter]

Codecov Report

Merging #509 (8462704) into master (c504c5f) will decrease coverage by 0.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #509      +/-   ##
==========================================
- Coverage   84.52%   84.51%   -0.02%     
==========================================
  Files          16       16              
  Lines        1409     1408       -1     
==========================================
- Hits         1191     1190       -1     
  Misses        218      218              

Impacted Files	Coverage Δ
src/promotion.jl	42.22% <ø> (-1.26%)	⬇️

---

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 c504c5f...8462704. Read the comment docs.

[giordano]

I don't think Quantity{T,S} is correct: the second parameter should be the dimension, not another numeric type. However, tests as you modified them pass for me without this method at all

[ChrisRackauckas]

Yeah, the better thing might be to completely remove this overload.

[giordano]

I'd argue it's a bit weird that in

julia> b = [0.0, 0.0u"m"]
2-element Vector{Quantity{Float64}}:
   0.0
 0.0 m

julia> b + b
2-element Vector{Number}:
     0.0
 0.0 m

b and b + b have different eltypes (I think that the proper eltype in both cases should be Number), but this is probably outside of the scope of this PR.