documentation Update AI transpiler tutorial for local mode, revised benchmarks, and Qiskit API updates

This PR overhauls the AI transpiler tutorial following the migration of the Transpiler Service and recent improvements in Qiskit’s standard transpiler stack. In addition to resolving issue #3631 , it updates the tutorial to work entirely in local mode, refreshes benchmarks and analysis, and aligns the notebook with the latest Qiskit APIs and documentation.

Key changes:

Converted tutorial to use local mode with generate_ai_pass_manager instead of the Transpiler Service
Updated circuit constructions: EfficientSU2 → efficient_su2, QFT → synth_qft_full, Permutation → PermutationGate
Removed outdated Quantum Volume benchmark
Added requirement: pip install 'qiskit-ibm-transpiler[ai-local-mode]'
Updated all descriptions to reference local AI passes instead of QTS
Switched backend to ibm_torino (133 qubits) and adjusted circuit sizes accordingly
Simplified workflow using the new transpile_with_metrics helper and updated benchmarking code
Reran all experiments and updated results, figures, and commentary to reflect current Qiskit performance
Removed ACG synthesis and added more circuits to produce more stable averages
Updated links and documentation references to match the current Qiskit release