Benchmark: solve each problem N times, take per-problem median

[Copilot]

The previous benchmark took the median over 11 distinct problem instances (one solve each), so timing variance came from problem difficulty differences rather than OS/scheduler noise.

Changes

• Two-level timing loop in benchmark_qp / benchmark_lp:

  • Outer loop generates num_problems (default 10) distinct problem instances
  • Inner loop solves each instance num_repeats (default 5) times; per-problem timing = median of those repeats
  • Reported solve_median = median of the 10 per-problem medians

• CSV schema: num_runs replaced with num_problems + num_repeats

for _ in 1:num_problems
    xref, H, f, A, bupper, blower, sense = generate_test_QP(...)
    prob_solve = Float64[]
    for _ in 1:num_repeats
        x, fval, exitflag, info = quadprog(H, f, A, bupper, blower, sense)
        push!(prob_solve, info.solve_time)
    end
    push!(solve_medians, median(prob_solve))  # smooth out scheduler jitter
end
# reported value: median(solve_medians)

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.