Fix detection of real cubic roots

[Liozou]

Thank you for this great package!

I encountered the following behaviour when looking at CO2 with a cubic equation of state:

julia> system = EPPR78(["carbon dioxide"]);

julia> Clapeyron.volume(system , 3311.0u"bar", 400.0u"K")
ERROR: UndefVarError: `vl` not defined
Stacktrace:
 [1] volume_impl(model::PR{…}, p::Float64, T::Float64, z::SVector{…}, phase::Symbol, threaded::Bool, vol0::Nothing)
   @ Clapeyron ~/.julia/packages/Clapeyron/JJGLA/src/models/cubic/equations.jl:269
 [2] volume(model::PR{…}, p::Float64, T::Float64, z::SVector{…}; phase::Symbol, threaded::Bool, vol0::Nothing)
   @ Clapeyron ~/.julia/packages/Clapeyron/JJGLA/src/methods/property_solvers/volume.jl:118 [inlined]
 [3] volume(model::PR{…}, p::Quantity{…}, T::Quantity{…}, z::SVector{…}; phase::Symbol, output::Unitful.FreeUnits{…})
   @ Clapeyron ~/.julia/packages/Clapeyron/JJGLA/src/methods/unitful_methods.jl:58 [inlined]
 [4] volume(model::PR{…}, p::Quantity{…}, T::Quantity{…}, z::SVector{…})
   @ Clapeyron ~/.julia/packages/Clapeyron/JJGLA/src/methods/unitful_methods.jl:55 [inlined]
 [5] top-level scope
   @ REPL[3]:1
Some type information was truncated. Use `show(err)` to see complete types.

Arguably worse, for a slightly different pressure, there is no error but the answer is obviously wrong (the volume is negative):

julia> Clapeyron.volume(system , 3363.1u"bar", 400.0u"K")
-5.070066466695018e-5 m^3

After investigating, it turns out that this is due to a faulty heuristic used to determine which roots of a cubic polynomial are real. This PR implements a cleaner heuristic that checks the value of the polynomial between the real parts of the roots, and compares the sign of these values to determine more precisely whether the roots are real. This yields more satisfying answers:

julia> Clapeyron.volume(system , 3311.0u"bar", 400.0u"K")
3.363139140634349e-5 m^3

julia> Clapeyron.volume(system , 3363.1u"bar", 400.0u"K")
3.3545657072357283e-5 m^3

And as a bonus, it seems that this bugfix incidentally makes the test for #172 pass (whereas it was marked as @test_broken).

[Liozou]

Nevermind the "bonus", it seems that CI is a bit flaky considering the last runs. Should I erase the @test_broken into @test conversion?

[longemen3000]

Nice!
Just comment that test out for now. Can you update the version number?, so we can release a new version as soon this is merged

[Liozou]

Sure, I just did that (and added a small comment by the test to explain what the issue is)!

[codecov]

Codecov Report

Merging #185 (76f4769) into master (45724c3) will decrease coverage by 0.01%.
The diff coverage is 100.00%.

❗ Current head 76f4769 differs from pull request most recent head 3b9eda7. Consider uploading reports for the commit 3b9eda7 to get more accurate results

@@            Coverage Diff             @@
##           master     #185      +/-   ##
==========================================
- Coverage   85.76%   85.76%   -0.01%     
==========================================
  Files         222      222              
  Lines       15099    15097       -2     
==========================================
- Hits        12950    12948       -2     
  Misses       2149     2149              

Impacted Files	Coverage Δ
src/models/cubic/equations.jl	85.27% <100.00%> (+0.59%)	⬆️
src/modules/solvers/poly.jl	100.00% <100.00%> (ø)

... and 11 files with indirect coverage changes