Add performance benchmark

[tmigot]

We can get inspiration from this: https://github.com/lanl-ansi/rosetta-opf/blob/main/nlpmodels.jl / https://discourse.julialang.org/t/ac-optimal-power-flow-in-various-nonlinear-optimization-frameworks/78486

---

First step is to have benchmarks to measure potential regressions:

• [x] Add benchmark folder
• [x] Pkg benchmark grad
• [x] Pkg benchmark jac
• [x] Pkg benchmark hess
• [ ] Pkg benchmark jprod
• [ ] Pkg benchmark jtprod
• [ ] Pkg benchmark hprod
• [ ] Add NLS variants
• [ ] Parallelize
• [ ] Add benchmark README
• [ ] Add PkgBenchmark CI
• [ ] Store benchmark result in the package
• [ ] Add documentation page listing available benchmarks

[gdalle]

Coming up in https://github.com/gdalle/SparsityDetectionComparison

[gdalle]

Here are the comparison results between JuMP, SCT and Symbolics for the first 90% or so of the OptimizationProblems.jl suite. A couple of the last problems take a really long time (probably for Symbolics) and I haven't yet narrowed down which ones. But I think the trend is clear.