Implement channel selection as intersection

The channel selection used to be a cascade of:

detection -> kurtosis -> peak time std -> init delay

However, removing channels between steps might reduct the preformance of some steps (especially peak time std). Now each step is independent and the output is the intersection between selected channels of the different filters.

@yuanx for the init_delay parameter, now the value is in ms. I think this is better because it can handle different sampling rates