Follow up on SpikeInterface/spikeinterface#4144

Currently, we get the MUX information (which ADC each channel was recorded from) from the ProbeTable and we assign it to each contact as a contact annotation. This is problematic when the user saves a subset of channels: in this case, the contacts are sliced and the MUX table is also sliced. BUT, we need the full MUX table to compute inter-sample-shifts (eg.g, if each ADC samples 12 channels, after slicing the contacts we don't know which of the 12 channels we are left with).

A couple of possible solutions:

1. compute the inter-sample-shift directly here.
2. in addition to storing the MUX channels, we should also store the indices of the channels sampled within a cycle (and then compute the inter-sample shift from there)

@chrishalcrow @samuelgarcia @h-mayorquin what do you think?