Fix t_starts not propagated to save_to_memory.

[JoeZiminski]

This PR closes #2515 by propagated t_starts to memory recording objects during save_to_memory. This is tested in #3117.

This works to propagate the times to save to memory, and tests are passing, but I havn't considered other uses for these recording classes. So it would be great if others could let me know if this might break anything elsewhere! I can't see why it would as it uses default kwargs that maintain the old behaviour, but still I'm paranoid 😅

Tests are added here that cover propagation of time information when stored as t_start and time_vector to various methods of saving the data. Also ,tests for other time-related functions (e.g. sample_time_to_index() are added).

[h-mayorquin]

Let's move the corresponding tests from #3117 to here.

I suspect that you tested this in #3117 because you wanted to have a canonical way for seting the t_start in your tests but you don't really need it, t_start is a public attribute of the segment class.

Plus, you are adding interface in #3117 and the tests should not depend on the newly added interface. Having t_start as an attribute of the segment class is a deper contract that the interface at the recording class and the tests of the interface should rely on the machinery that is already baked in on the data model.

[JoeZiminski]

Thanks both, I will move the tests over now

[JoeZiminski]

@h-mayorquin @samuelgarcia @alejoe91 I re-requested review as added tests from #3117 to here, these cover all saving behaviour for both time representations + some other time-related functions. Herberto I tried to reduce the indirection on the fixtures by a level, let me know if it is still not clear.

[alejoe91]

LGTM!!

Removed the deprecated tests ;)

tests were added