Optimise playback of static meshes in Viewer

[bentoogood]

Playback of complex static scenes is slow - probably as gaffer isn't aware that it is unchanging frame to frame.

Make some optimisations by detecting static geo and avoid regenerating each frame.

Related to #121

[johnhaddon]

This relates specifically to the implementation of SceneReader::hash(), which is currently assuming that the whole scene depends on time. We need to detect static geo and not append time to the hash.

[johnhaddon]

We also need to ensure that time doesn't affect the hash for the child names as it currently does - SceneCache hierarchies are constant across time.

I believe there is also currently significant overhead in the SceneHierarchy ui which is rebuilding every frame - this seems unnecessary given that the hierarchy itself rarely changes from frame to frame (it could, but still, what's the use of displaying that while someone's playing back?). So I think with some sort of playbackStarted() and playbackStopped() signal we could disable updates temporarily for the SceneHierarchy. This would probably mean making a Playback class, or adding playback methods to the ScriptNode, at which point we'd be in a good position to flesh out Ben's pull request #559.

[johnhaddon]

We might also want to investigate the ideas in #427 in conjunction with this ticket, as I think they could speed up scene evaluation a fair bit across the board.

[johnhaddon]

There's some initial work on this here :

https://github.com/johnhaddon/gaffer/tree/sceneReaderOptimisation

[johnhaddon]

OK, now this ticket is a bit more a catch-all for performance improvements, here's a checklist of what needs doing before we can consider it closed :

• Static geo hash optimisation in SceneReader
• ScenePath plugDirtiedSignal() lazy connections
• SceneHierarchy disabling update during playback
• Optimise StringPlug storage to avoid string substitutions when there are no variable references
• Other things that come up during investigation

[johnhaddon]

I think we've made enough general playback optimisations to call this one good now - greater than 5x speedup overall for the bunny scene. This can be closed when we get code reviews for each of the individual pull requests done.