Performance: Limit data layer state updates

[torkelo]

Various performance fixes

•  We were using infinity replay subjects so when coming back to a scene data layers emitted all their previous emitted results
• Ignore replayed data layer results that has already been processed (this happens when a data provider get's re-activated, we subscribe the the SceneDataLayerSet this causes the ReplaySubject to emit last data, even if that has already been incorporating into the current state). Created a new DataLayersMerger that can detect when a result has already been seen.
• In SceneQueryRunner ignore updating state when data layers are received if there are no annotations (And the previous result also had no annotations)

📦 Published PR as canary version: 4.22.0--canary.724.9079775010.0

✨ Test out this PR locally via:

npm install @grafana/scenes@4.22.0--canary.724.9079775010.0
# or 
yarn add @grafana/scenes@4.22.0--canary.724.9079775010.0

[dprokop]

This condition is quite hard to reason about, as it assumes that one knows that the alert state is a reference to an state taken from the alert states results. I wonder if it wouldn't be easier if we didn't set that particular alert state reference in the state but always create a new object, and here instead rely on object equality rather than reference equalit? WDYT?

[torkelo]

@dprokop pushed update that switched to isEqual instead, for some reason I thought alertState was an enum :)

[dprokop]

Gave it another look (thanks for updating the alerts state quality check!) and it looks good. I've also tested some test dashboards with annotations and it works great.

[grafanabot]

🚀 PR was released in v4.22.0 🚀