Test parallel connected-app writes across two collab replicas

[HMarzban]

Parent

#328. Follow-up to #229.

Problem

The #229 fix serializes writes with a per-document lock and waits until the save commits (hocuspocusApply.ts withDocumentLock). The lock lives in one process. Production runs two hocuspocus-server replicas behind HOCUSPOCUS_INTERNAL_URL (docker-compose.prod.yml), so two parallel writes to one cold document can reach two appliers with two locks. The clean result in #229 was measured on one process only. Writes in sequence are safe, because a cold 200 waits for its row.

What to do

Measure before changing anything. Run two collab processes on two ports, the way #226 tested two replicas. Send five parallel edit_blocks calls to one section of a cold document with the same rev, split across both processes. Then read the document.

Acceptance

• The result is posted here: the final text, and the answer each call got.
• If exactly one write applies and the rest answer a conflict or busy, close this issue.
• If text is lost or doubled, post the failure and propose the smallest fix here before building it.

Out of scope

Any change to the lock before the measurement.