✨ Collect WebSocket connections opened before init()

[bdibon]

Motivation

WebSocket collection subscribed to initWebSocketObservable() only once init() had run, so any
connection opened before then was invisible to RUM. A socket that a page opens at load — the common
case for chat, live feeds and collaborative editors — was never reported.

Changes

• startBufferingData() gains a BufferedDataType.WEB_SOCKET source, alongside runtime errors,
  fetch, XHR and console. Instrumentation now starts at SDK load rather than at init().
• The subscription is unconditional: the WebSocket opt-in is only knowable at init(), and
  consulting it before instrumenting would miss exactly the connections this change is about.
• webSocketCollection consumes bufferedDataObservable instead of the raw
  Observable<WebSocketContext>, so pre-init() activity is replayed on unbuffer.
• The opt-in gate (trackResources plus betaTrackWebSockets or the TRACK_WEBSOCKETS
  experimental feature) moves from startRum into startWebSocketCollection. startRum calls it
  unconditionally; when the gate is closed nothing is subscribed or allocated and the returned
  stop is a no-op.
• startBufferingData() now takes an optional list of BufferedDataType sources to instrument.
  makeLogsPublicApi passes only RUNTIME_ERROR, FETCH, XHR and CONSOLE — logs never
  instruments WebSocket or pays its buffering cost. makeRumPublicApi passes all five sources,
  including WEB_SOCKET.
• Tests: the fake WebSocket inlined in the observable spec becomes a shared core test utility
  (mockWebSocket), and the collection specs now drive through instrumentation rather than pushing
  synthetic contexts into an observable.

Known issues to resolve before merge

Two problems were surfaced while building this and deliberately deferred — each gets its own PR:

1. WebSocket traffic can exhaust the buffered observable during pre-init. BufferedObservable
   is a 500-entry ring shared with runtime errors, fetch, XHR and console. message-in /
   message-out fire per message, so one chatty socket over a seconds-long pre-init() window can
   evict every other buffered entry. Nothing is coalesced here; the existing
   'Early data collection dropped data on unbuffer' debug telemetry is the tripwire. Coalescing
   (and where the aggregation should live) is addressed in a distinct PR.
2. No e2e tests exercise the pre-init behavior. ✅ Add E2E coverage for WebSocket collection before init() #4939 introduces a TestBuilder.withPreInitScript() method to delay the init phase and execute code during the pre-init window.

Test instructions

1. yarn dev, then in sandbox/index.html open a WebSocket before DD_RUM.init() and enable
   the opt-in on init:

   const ws = new WebSocket('wss://echo.websocket.org')
   ws.addEventListener('open', () => ws.send('hello'))
   // ...existing DD_RUM.init({...}) call, plus:
   //   trackResources: true, betaTrackWebSockets: true

2. Load the page and let the socket close (or call ws.close() from the console).
3. Confirm a websocket resource event is sent with the connection's message counts and sizes —
   the pre-init() connection is reported rather than dropped.
4. Reload with betaTrackWebSockets removed and confirm no WebSocket resource event is sent.
5. Confirm a logs-only page (DD_LOGS.init() without RUM) never patches the global WebSocket.

Checklist

• Tested locally
• Tested on staging
• Added unit tests for this change.
• Added e2e/integration tests for this change.
• Updated documentation and/or relevant AGENTS.md file

[cit-pr-commenter-54b7da]

Bundles Sizes Evolution

📦 Bundle Name	Base Size	Local Size	𝚫	𝚫%	Status
Rum	180.44 KiB	180.58 KiB	+142 B	+0.08%	✅
Rum Profiler	8.43 KiB	8.43 KiB	0 B	0.00%	✅
Rum Recorder	21.12 KiB	21.12 KiB	0 B	0.00%	✅
Logs	57.04 KiB	58.78 KiB	+1.74 KiB	+3.05%	✅
Rum Salesforce	N/A	138.59 KiB	N/A	N/A	N/A
Rum Slim	138.47 KiB	138.59 KiB	+121 B	+0.09%	✅
Worker	22.96 KiB	22.96 KiB	0 B	0.00%	✅
Rum Shopify	N/A	200.16 KiB	N/A	N/A	N/A
Rum-shopify Profiler	N/A	8.43 KiB	N/A	N/A	N/A
Rum-shopify Recorder	N/A	3.72 KiB	N/A	N/A	N/A

[datadog-datadog-prod-us1-2]

🎉 All green!

🧪 All tests passed
❄️ No new flaky tests detected

🎯 Code Coverage (details)
• Patch Coverage: 87.27%
• Overall Coverage: 77.26% (+0.06%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: fa87695 | Docs | Datadog PR Page | Give us feedback!}