Cross-Context Backdoor Attacks against Graph Prompt Learning