✨ [abr_sim] Add warm-up phase support in ABR simulation

Introduce a warm-up phase to ABR simulation that is excluded from test
metrics. This allows for stabilization before measuring performance.

Updated `run_rule_policy` and `run_llm` functions to support warm-up
steps, adjust max episode steps, and log warm-up events separately.
Added CLI argument `--warmup-steps` for configuration.

Author: YuanmingLeee