BitLoRA: Quantization-Compatible Adapter Tuning for 1.58-bit LLM in Federated On-Device AI-Agent | Synapse