DTMM: Deploying TinyML Models on Extremely Weak IoT Devices with Pruning