mapping between gtk timestamp and instant

[Doublonmousse]

Fixes #1289

I'm not sure if I need to take into account the overflow, this complicates things quite a bit.
It's a little bit ugly, but it's probably enough to test things out.

I'm getting a lot of 0 for event times when scrolling, I need to see what event triggers this so there isn't any residual weirdness.

Well, it seems every single scroll event will trigger a MotionNotify without a timestamp.

[Doublonmousse]

More weirdness :

• two events can be sent with the same gtk time (hence why the backlog policy is useful, even if it's to put it at 1 ms)
• scroll events don't have timestamps (I think that would only affect things when scrolling with the left click pressed)

[Doublonmousse]

After thinking a bit, I'm not sure if I'm not making a bit deal of a (relatively) small issue. We have a timestamp for all events, albeit sometimes less precise than gtk when it gives it, with larger deviations in the case of the ButtonPress for pen on windows. But for all time sensitive algorithm, as long as we do the following

• reject negative time deltas
• robustness to time deltas that are too small (compared to their real duration)
  Things will be fine.

Maybe I can check on my hardware by how much doing the Instant::now vs relying on gtk time changes things