Pre-Compiled Pipeline Shards for Distributed LLM Inference on Intel AI PC Fleets