Performance benchmark

[mtanneau]

I'll write a benchmark script before working on performance improvements.

I propose to use 3 benchmark sets:

• The Netlib LPs (available here): these are small LPs, in .SIF files and fixed MPS format.
• The Maros-Meszaros QPs (here)
• The LPs of Hans Mittelmann's benchmark (see here). These are (much) larger LPs in .mps, some with integer variables, and some in free MPS format.

The reason for including a larger testset of instances would be to allow modifications that improve performance on large instances, while allowing a limited degradation on smaller instances.

For information, the readers in C/C++ solvers (commercial and open-source) are ~10-30x faster than the ones we have in Julia.

[dpo]

Sounds like a plan! It would be great to see some flame graphs in SVG to see what the bottlenecks are.

[mtanneau]

Closed by #30