Refactor pool of solutions for DwSp & insertion of a column

[guimarqu]

Changes

1. Move the pool of primal solutions from the Formulation to the DwSp duty because it's specific to pricing problems.

2. Refactor column insertion.
   Before, we had to call two methods to insert a new column in the master formulation. These two methods were always called together.

var_was_inserted, sol_id = ClMP.setprimalsol!(cbdata.form, sol)
if var_was_inserted
    mc_1 = ClMP.setcol_from_sp_primalsol!(
        masterform, cbdata.form, sol_id, string("MC_", ClA.getsortuid(sol_id)), ClMP.MasterCol
    )
else
    mc_1 = ClMP.getvar(masterform, sol_id)
end

Now, one method does the job.

_, col_id = ClMP.insert_column!(masterform, sol, "MC")

Consequence

In column generation, the insertion of a column in the pool is not parallelized anymore.

it's not possible to parallelize column insertion because of concurrent access to the master formulation. Therefore, we created setprimalsol! method to insert the solution in the pool because we wanted to parallelized insertions of solutions in the pool of subproblems. However, I think that we do not gain a lot of performance because Coluna spends most of the time solving the subproblem. Since the tradeoff performance/code simplicity is in favor of the latter, I think we can drop parallelized insertion.

About the data structure of the pool

My initial idea was to replace the dynamic sparse matrix that stores the pool with a dict of dict.
This is more complicated than I thought.

First, we use column and row views (i.e. variables used by a given column and columns that contain a given variable). It makes the change more complicated because we need to maintain two data structures, a col-major one and a row-major one.

Second, I thought Dictionary was supporting Sorted Dict. Actually, it's an OrderedDict that preserves only the order of insertion. We may use a SortedDict from DataStructures but it does not implement math operations with arrays.

So I suggest to not change the data structure for now and improve the performance of DynamicSparseArrays.

[codecov]

Codecov Report

Merging #609 (b4f137b) into master (a4d23e1) will increase coverage by 0.00%.
The diff coverage is 87.20%.

@@           Coverage Diff           @@
##           master     #609   +/-   ##
=======================================
  Coverage   86.24%   86.25%           
=======================================
  Files          48       48           
  Lines        5047     5057   +10     
=======================================
+ Hits         4353     4362    +9     
- Misses        694      695    +1     

Impacted Files	Coverage Δ
src/Coluna.jl	100.00% <ø> (ø)
src/MathProg/manager.jl	61.76% <ø> (ø)
src/Algorithm/colgen.jl	83.54% <70.00%> (-1.25%)	⬇️
src/MathProg/formulation.jl	78.59% <98.24%> (+0.39%)	⬆️
src/Algorithm/preprocessing.jl	76.68% <100.00%> (-0.25%)	⬇️
src/MOIwrapper.jl	85.48% <100.00%> (ø)
src/MathProg/clone.jl	100.00% <100.00%> (ø)
src/MathProg/duties.jl	69.23% <100.00%> (+2.56%)	⬆️
src/MathProg/projection.jl	95.23% <100.00%> (ø)
src/MathProg/vcids.jl	81.48% <100.00%> (+2.31%)	⬆️
... and 6 more

---

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 a4d23e1...b4f137b. Read the comment docs.