Bus enricher

ReactionBus runs a single enricher before any subscriber receives a Reaction. The enricher resolves the audio clip URL + duration, calls FaceLibrary.selectIndices(count:) to deduplicate face picks across multiple displays, and stamps publishedAt. The output is a FiredReaction. Audio, screen flash, LED, and notification all see the same enriched envelope, so a reaction stays in lockstep regardless of how many outputs are firing.

Before any subscriber receives a reaction, ReactionBus runs a registered enricher closure exactly once. The enricher resolves all per-reaction metadata so that every subscriber operates on identical, pre-computed values. The bus stamps publishedAt at the moment publish(_:) is called. before enrichment begins. so the timestamp is stable regardless of enrichment latency. If enrichment takes longer than 500 ms, the bus publishes a minimal fallback FiredReaction.

The enricher, registered by Yamete.startOutputs(), does three things:

• Calls AudioPlayer.peekSound(intensity:reaction:) to dedup-filter and intensity-select an audio clip URL and duration. For non-impact reactions, peekSound returns nil and the bus falls back to ReactionsConfig.eventResponseDuration.

      - Calls `FaceLibrary.shared.selectIndices(count:)` with the current connected screen count to pick one face index per display, scored for recency dedup.
  
      - Returns a `FiredReaction` carrying the resolved `soundURL`, `clipDuration`, `faceIndices`, and `publishedAt`.