Type-generic S matrix in LDL factorization

[matbesancon]

Still needs tests, addresses in part the issue observed in #64

[codecov-io]

Codecov Report

Merging #76 (dd9913e) into master (039132b) will decrease coverage by 0.00%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master      #76      +/-   ##
==========================================
- Coverage   89.90%   89.89%   -0.01%     
==========================================
  Files          36       35       -1     
  Lines        2526     2524       -2     
==========================================
- Hits         2271     2269       -2     
  Misses        255      255              

Impacted Files	Coverage Δ
src/KKT/ldlfact.jl	94.11% <ø> (ø)
src/KKT/KKT.jl	77.77% <0.00%> (-2.23%)	⬇️
src/Tulip.jl

---

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 039132b...e3f82e0. Read the comment docs.

[matbesancon]

I'm checking whether it passes to also add Float32 to the Tv types in the tests. This may be overkill because repeating all tests a third time.

[matbesancon]

The Float32 addition adds a lot of tests, up to you to see whether you want to keep them. I believe they are interesting because they prevent bugs for types that would get absorbed by promotion with Float64 (not the case for BigFloat)

[mtanneau]

Thanks!

I don't mind running the tests with more types. It's the MOI tests that take most time; currently they only run in Float64.