Single-slot sync plus UI-thread snapshot apply can drop a fire-and-forget command #622

Problem

Poll, spine, full-state, and one-shot work contend for one sync slot. Fire-and-forget callers depend on retry conventions, so a new command can be lost if it does not re-arm. Snapshot projection and structural updates still run through one UI-thread apply graph.

Not #509. #509 is the 197 ms participant-rebuild stutter. Same apply graph, different failure: lost commands versus jank.

Evidence

Static review at 6f4f025 (2026-09-24).

• MediaCoreSupervisor.cs ~670-672
• MediaCoreBridgeService.cs
• StudioViewModel.cs snapshot/sync paths

Done when

• Production sync has one scheduler with a stated contention policy (queue, coalesce, or reject-visible).
• A one-shot command that loses the slot is retried by the scheduler or failed to the caller. Silent drop is not allowed.
• A test constructs two overlapping syncs and proves the second command is not discarded.

Related, not duplicate

• Operator stutter: what flips the participant structure signature, and which rebuild step costs 197ms #509 operator stutter / 197 ms rebuild
• Split ingest off MediaCore.cpp after F1 has one consumer #540 ingest split off MediaCore