Some improvements to time handling

As discussed in #3509, some improvements to the time-handling could be made:

1. get_times() returns a view to the segment time_vector if it exist. This means that changing the returned array will also change the one stored on the segment. This is different behaviour to if .t_start is used under the hood, as the array will be generated on the fly. I think everyone is agreed that get_times() should return a copy of the time_vector (if it exists).

2. copy.deepcopy(recording) or I think recording.clone() does not copy the time_vector. This is discussed here. An fix could be:

What about times for clone()? This is not handled at the BaseExtractor level but the BaseRecording could override it or add an clone_extra_stuff() in base class and the BaseRecording could just implement it

Start by tracing get_times() and time_vector in BaseRecording and related BaseExtractor clone or deepcopy handling; issue #3509 contains the referenced discussion. Confirm that get_times() does not expose the stored array and that cloning preserves time_vector, then add or update the relevant tests if the existing test locations are found.