Rename CLI flags to be consistent between train.py and inference.py

[DorisMai]

1. --geometry_cache_name → --geometry_cache (or vice versa) — unify across both scripts
2. --rk4_steps → --num_steps in train.py to match inference.py
3. --use_sc → --use_self_cond in inference.py to match train.py
4. Add --log_level and --log_file to inference.py for consistent logging control, also remove redundant setup_logging_for_tqdm() at the beginning of inference.py script (related to copilot comment)