Script for benchmarking

Problems to take into accounts:

• Input for pretrained detectors to shorten (at least for XLMR) <512 tokens (due to pretraining) - cf MR#26