HYDRA-1457 : Fix UsdStageMap::allStages to not rely on the internal cache when it can be outdated

[debloip-adsk]

This fix is notably required in order to get some UFE camera changes working in Maya.

[pierrebai-adsk]

IDK if there can be perf implication of calling getAllNames(), but we'll see.

[seando-adsk]

IDK if there can be perf implication of calling getAllNames(), but we'll see.

I think there is which is one of the reasons we are using the stage cache here. @debloip-adsk It seems to me like the "real" problem could be a missing call to setDirty() to make sure that when allStages() is called it will rebuild the cache first.

[ppt-adsk]

Are there any performance implications for this change? How often is this run?

[seando-adsk]

Are there any performance implications for this change? How often is this run?

That is kind of hard to say. Its available from a python function called getAllStages() and I've seen it being called in a few code places, such as from SceneSegmentHandler::findUsdGatewayItems() and the proxyRenderDelegate.

[debloip-adsk]

The core problem I'm seeing here is : when/from where would we set the cache dirty? Once the node rename happens, there is a UFE notification and a Maya notification sent, but Maya UFE camera code receives the UFE notification first, and then immediately queries MayaUsd's UsdStageMap through Ufe::SceneSegmentHandler::findGatewayItems before it has received the Maya node rename notification. So MayaUsd has no idea any change has happened before it is queried for updated info. Even if we added support in MayaUsd's UsdStageMap for handling UFE notifications, we'd still depend on the notification order, which doesn't seem super robust either. Is there a way to set the cache dirty through Maya/UFE?