feat: add OptimizerBatchLocalSearch

[sumny]

Things to discuss:

• [x] check_values in the objective but also instance still induce high overhead in general in bbotk
• [x] got massive speedups (factor 4-5) by not relying on active bindings such as class, lower, upper, levels, etc. of search space which still hints at issues with paradox
• [x] rename mu, n_points_initial_random, n_points, sigma?
• [x] in mbo config, check if local search actually runs correctly; think about warm starting AcqOptimizer

Things to do:

• [x] test on mixed spaces and mixed hierarchical spaces
• [x] benchmark on some standard test functions
• [x] discuss n_initial_points_random, mu and n_points
• [x] adjust example and n_evals

[sumny]

Results of running random search vs. local search with a budget of 1000 + 10000 evaluations on 2D Branin. Mean log regret to global best (0.3978874) over 30 replications, ribbons are standard errors.