Stalls that remain:

• Families, the biggest one. Family registration, bind and unbind still hold the engine loop for the whole partition hydrate, and nothing bounds that hydrate. With queryFamilies on by default, that covers every query with an equality filter at the root, so for those queries the original stall is fully there. This is 429 §6.1.1, planned but not built. It needs a per-partition rollback first.
• Writes on the same worker. A commit still queues behind a cold hydrate, for up to one hydrate budget (5 s by default). My P1 fix changes what happens behind a pile of reads: the writer now waits instead of the worker being reaped. It doesn't shorten the wait. Removing it needs change (B) of the snapshot-pin design, the double-buffered snapshot.
• Other commands on a worker that is mid-hydrate. The loop still waits, for up to one budget, when it sends anything that expects a reply to a busy worker, such as a second registration landing on the same worker.