What is the purpose of blocking the return until all requested key frames have been generated? This was previously required/done to get "a" timestamp but no longer serves any purpose?

See also #145 on the overall broken assumptions

This issue had an associated resolution in WebRTC September 2025 meeting – 16 September 2025 (PR #276: Default the generate key frame algorithm to all layers):

RESOLUTION: proceed with merging PR

Note that RIDs are optional, we only know that either all encodings have RIDs or none of them have RIDs, but we don't know that no RIDs equals singlecast - you could have multiple encodings without RIDs. The alternative would be encodingIndex but we can say that rids are a prerequisite

I've now updated the algorithm to work sans rids. PTAL @henbos @youennf

Note I also cleaned up the language a bit:

"encoder" was being used at two different conceptual levels: A) as the owner of all layers, and B) per layer. I reworked/replaced the latter with "layer" consistent with existing text in 4.10: "When transform’s [[encoder]] receives a keyframe request, ... Set rid to the RID of the appropriate layer, or undefined if the request is not for a specific layer."