fix: bound y-websocket reconnect rate on transient post-open closes

[kptdobe]

Summary

• y-websocket resets its wsUnsuccessfulReconnects counter on every successful onopen, so any close that follows a brief successful handshake reschedules at ~100 ms. The 4401/4403 guard from fix: block y-websocket auto-reconnect during async IMS refresh #943 does not cover non-auth paths (1011 from initSession catch, 1005 from closeConn, 1006 from socket reset), and single users behind corporate Zscaler proxies have sustained 5k+ WS upgrades/sec/IP because of it (see COR-43).
• Add a rapid-reconnect guard to createConnection: track open-then-close intervals on the provider and apply manual exponential backoff (1s/2s/4s/… capped at SHORT_SESSION_MAX_MS = 30 s) when sessions shorter than MIN_HEALTHY_SESSION_MS = 5 s repeat. A healthy session (≥ 5 s lived) resets the counter so a routine reconnect after a long session still reconnects via y-websocket's own timer.
• The auth path from fix: block y-websocket auto-reconnect during async IMS refresh #943 is unchanged and the new guard never increments on 4401/4403 closes — the auth flow has its own loop guard via lastSentToken.

Implementation notes

• The new constants live as module-private consts in blocks/edit/prose/index.js.
• The guard sits BEFORE the existing token-refresh provider.protocols reassignment so a token refresh still happens for the next manual provider.connect() after the backoff.
• y-websocket's internal 100 ms setTimeout(setupWS) still fires from onclose, but provider.disconnect() flips shouldConnect = false, so that timer's setupWS call no-ops. The manual setTimeout is what re-arms the connection.
• No changes to da-y-wrapper or upstream y-websocket — the fix sits at the consuming-call site so it ships with da-live's normal release cadence.

Test plan

• npm run lint clean on the touched files
• Focused unit tests pass: npx wtr "./test/unit/blocks/edit/prose/index.test.js" — 35/35 green, including the new prose/index createConnection rapid-reconnect guard (COR-44) suite (7 new scenarios) plus an updated existing test that now reflects the post-guard shouldConnect = false parking state.
• After merge, query Coralogix and confirm the dominant da-collab URLs from the COR-1 daily review drop out of the >95% exception-ratio bucket and that the populated "Network connection lost." exception count returns to baseline (<10k/24h) within 48 h.

Test: https://cor44--da-live--adobe.aem.live/

Refs

• Issue: COR-44 (rapid-reconnect guard)
• Parent: COR-43 (investigation that surfaced the y-websocket reset behavior)
• Prior PR: fix: block y-websocket auto-reconnect during async IMS refresh #943 (auth-close 4401/4403 guard)
• Supersedes: fix: bound y-websocket reconnect rate on transient post-open closes #954 (same change on a non-compliant branch name)

🤖 Generated with Claude Code

[aem-code-sync]

Hello, I'm the AEM Code Sync Bot and I will run some actions to deploy your branch.
In case there are problems, just click the checkbox below to rerun the respective action.

• Re-sync branch

Commits

• 17423a0 ✅ (latest)
• 9664da7 ✅
• 1f5556e ❌