Optimization of get_noise_levels (and thus get_random_data_chunks)

[yger]

Actually, get_noise_levels makes internal use of get_random_data_chunks that is not parallel. There is thus room for improvements, via parallelization. This could be achieved at low cost by using the run_node_pipeline machinery, assuming some patch like #2011 would be accepted. If this is the case (or we can discuss somehting in line and similar), one could simply provide a list of fake "peak" times, some temporal window, and use the WaveformDense object of the pipeline to get data in parallel. Adding an extra job_kwargs to get_noise_levels would then speed up everything. @samuelgarcia what do you think?