fix(ember): Stop retaining component render payloads in beforeEntries

[yashschandra]

Before submitting a pull request, please take a look at our
Contributing guidelines and verify:

• If you've added code that should be tested, please add tests.
• Ensure your code lints and the test suite passes (yarn lint) & (yarn test).
• Link an issue if there is one related to your pull request. If no issue is linked, one will be auto-generated and linked.

Problem

_instrumentComponents subscribes to Ember's render.component (and optionally render.getComponentDefinition) instrumentation and stores { payload, now } into a module-scope beforeEntries map keyed by payload.object (a per-component-instance GUID). processComponentRenderAfter reads the entry but never removes it, so every instrumented component render is retained for the lifetime of the page.

Since the render payload references the component instance (view), this pins each destroyed component and everything it references. In our Ember app (Customer.io Journeys) this retained every destroyed message-composer graph when users switched between wizard steps — ~34 MB of unreclaimable heap per step switch, measured in Chrome with --js-flags=--expose-gc and confirmed via heap-snapshot retainer paths ending at the instrumentation subscriber's beforeEntries closure:

(GC roots) → module context → subscribers → before closure → beforeEntries
  → entry → payload → view (component) → parent component graph (~34 MB / cycle)

Fix

• Convert RenderEntries to a WeakMap keyed on the render payload object itself (per review feedback). This guarantees an entry can never outlive its payload, even if a render's after hook is never called. Ember's @ember/instrumentation passes the same payload object to a subscriber's before and after hooks, so identity-based lookup is safe; payload.object (a string) can't key a WeakMap, so the entry no longer needs to store the payload at all.
• Still delete the entry in processComponentRenderAfter once the render completes, so a long-lived payload doesn't hold a stale entry between renders.
• Export _processComponentRenderBefore / _processComponentRenderAfter (following the existing _-prefixed pattern used by instrumentEmberAppInstanceForPerformance) and add unit tests covering: entry removal on completion (both below and above minimumComponentRenderDuration), and that an after without a matching before leaves other in-flight entries untouched.

Verified in the affected app: with this change all destroyed component graphs become collectable and back-to-back heap snapshots across step-switch cycles show only ~0.5 MB/cycle of JIT-code growth instead of ~34 MB/cycle of retained objects.

🤖 Generated with Claude Code

[github-actions]

👋 @nicohrubec, @s1gr1d — Please review this PR when you get a chance!

[s1gr1d]

L: Maybe it would make sense here to use a WeakMap and use the payload as the key. We don't need the info, when the payload is already garbage collected, right?

[mydea]

👍 I think moving this to a weakmap is likely the easiest solution here?

[mydea]

(I think it's still ok to remove the entries as well, but a weak map ensures nothing can linger around if something is not called etc)

[yashschandra]

Done