Simpler depth bounded mapper
The depth bounded mapper contained some code to place local gates first. But the depth bounded algorithm will generally wants to minimize the cost of the most expensive gate. Placing local gates first may interfere with this.
To compare, I've run random circuits of the old depth-bounded algorithm versus the new one (v2). In most cases v2 performs better, especially on asynchronous cost, but on depth it seems worse at times. Below are the results:
random_file_grid_qubits_vs_cumulative_cost.pdf
random_file_grid_qubits_vs_synchronous_cost.pdf
random_file_grid_qubits_vs_depth.pdf
random_file_modular_qubits_vs_cumulative_cost.pdf
random_file_modular_qubits_vs_depth.pdf
random_file_modular_qubits_vs_synchronous_cost.pdf
random_file_path_qubits_vs_asynchronous_cost.pdf
random_file_path_qubits_vs_cumulative_cost.pdf