Batch kernel tracker

[klamike]

This issue tracks what kernels are left to implement the batched version of (ref #78 which started this work by batching factorize_regularized_system! and solve_system!). Roughly, the functions currently taking the most time are at the top.

• check_residual! aka post_solve! (waiting for batched mul!(::KKTSystem) Don't call check_residual! in batch klamike/MadIPM.jl#2 (comment))
• evaluate_model! - batch/parametric NLPModels API?
• update_termination_criteria! (dual_objective, get_inf_pr, get_inf_du, get_optimality_gap)
• set_predictive_rhs
• set_correction_rhs
• apply_step!
• MPCSolver allocation/sharing buffers -- MK working on this now
• pre_initialize!
• post_initialize!
• prediction_step_size! (prediction_step_size! batched klamike/MadIPM.jl#3)
• set_aug_diagonal_reg! (set_aug_diagonal_reg! batched klamike/MadIPM.jl#4)
• update_step_size! (Batched update_step_size! klamike/MadIPM.jl#5)

Comments left to address:

• What should the logs look like? Add support for solving uniform batches with CUDSS #78 (comment)
• Properly handle linear solver options with NoLinearSolver Add support for solving uniform batches with CUDSS #78 (comment)
• Debug kwargs API in solve! Add support for solving uniform batches with CUDSS #78 (comment)
• Global options instead of using first solver (FIXME comments)
• Update to MadNLP 0.9 upon release (might include async change)

[klamike]

1354_pegase x 16: First a loop over madipm, then the uniform batch solve:

Zooming in on one step of the uniform batch solve:

[klamike]

This approach has been abandoned in favor of #91