benchmark on trajectory games

[dfridovi]

• factor out benchmarking code to work for different types of problems (i.e., adding BenchmarkType abstraction)
• introduce TrajectoryGameBenchmark modeled on LaneChange in examples/

[dfridovi]

Based on these results, I think we need to roll back the Krylov stuff, or at least make it not the default option.

julia> data = SolverBenchmarks.benchmark(SolverBenchmarks.TrajectoryGameBenchmark(); num_samples=10); SolverBenchmarks.summary_statistics(data)
[ Info: Generating random problems...
[ Info: Generating IP MCP...
[ Info: Generating PATH MCP...
[ Info: Warming up IP solver...
[ Info: Warming up PATH solver...
Solving IP MCPs... 100%|█████████████████████████████████| Time: 0:00:53
Solving PATH MCPs... 100%|███████████████████████████████| Time: 0:00:02
[ Info: IP runtime is 2918.485214038064 % that of PATH.
(ip = (success_rate = 0.9, μ = 5.099074351777778, σ = 0.3451189449072598), path = (success_rate = 0.7, μ = 0.17471647028571427, σ = 0.08759241497713036))