On Feasibility of Server-side Backdoor Attacks on Split Learning