Better timing

[lukassnoek]

Implement a new, more robust timing system:

• first window flip is synchronized to start of experiment;
• onset of phases are now actually synchronized to the window flip (not before the flip, so now timing is up to 1 frame more accurate than before);
• Trial objects have a parameter timing, which can be set to "seconds", in which phase_durations are interpreted as seconds, and "frames", in which phase_durations are interpreted as number of frames. The latter option is more precise, given that you're not dropping frames.
• Also keeps track of number of window flips/frames that a phase is shown for;

Also changed the logging system:

• There is no "log" per Trial, only at the Session level (.log attribute); this is because durations are determined after all trials are finished (i.e., durations = diff(onsets)).

[lukassnoek]

Apologies, should be a PR to the develop branch