feat: adopt Cloudflare WebSocket Hibernation API

[kptdobe]

Problem

During image drag-and-drop sessions in DA, Durable Objects are evicted when the browser briefly drops WebSockets (while busy uploading blobs). Each eviction clears the in-memory docs Map. On reconnect, bindState re-fetches from da-admin and triggers a 1-second revert timer, causing the just-inserted images to disappear.

Root cause

webSocket.accept() (the non-Hibernation API) causes the DO to be evicted whenever a WebSocket disconnects. Any work in flight (debounced saves, bindState async chain) is cancelled with Outcome: "canceled", and the in-memory Yjs state is lost.

Solution

Adopt the Cloudflare WebSocket Hibernation API (ctx.acceptWebSocket()), which keeps the DO alive across WebSocket disconnects:

• DocRoom.fetch() calls this.ctx.acceptWebSocket(server) + server.serializeAttachment({ docName, auth, authActions }) instead of server.accept()
• New class methods webSocketMessage, webSocketClose, webSocketError handle CF Hibernation routing
• setupWSConnection gains a hibernation flag (default false) — when true, skips addEventListener registration since CF calls the class methods directly
• New exports handleWebSocketMessage and handleWebSocketClose in shareddoc.js allow webSocketMessage/webSocketClose to re-hydrate Yjs state after a cold start

Tests

• Updated 4 existing WebSocket tests to use ctx.acceptWebSocket mock + serializeAttachment
• Added 6 new tests: cold-start re-hydration, warm-start (no redundant bindState), webSocketClose cleanup, webSocketError cleanup, no-op close on unknown doc, read-only auth restoration

🤖 Generated with Claude Code

[bosschaert]

In theory this seems like a good idea. I don't have any experience with the hibernation API so we need to make sure that we test it very well.

@karlpauls any thoughts?

[kptdobe]

Need more testing, moving to draft PR.