fix(input): clear PointerEventReceiver events independently of PointerSystem (#3356)

[MohammadYusif]

===:clipboard: PR Checklist :clipboard:===

• 📌 issue exists in github for these changes (PointerEventReceiver memory leak #3356)
• 🔬 existing tests still pass
• 🙈 code conforms to the style guide
• 📐 new tests written and passing / old tests updated with new scenario(s)
• 📄 changelog entry added (or not needed)

==================

Closes #3356

Changes:

PointerEventReceiver accumulates native pointer events in its currentFrameDown / currentFrameUp / currentFrameMove / currentFrameCancel / currentFrameWheel arrays. Those arrays were only ever consumed (update()) and freed (clear()) by PointerSystem.update() each frame. If a scene removes the PointerSystem (e.g. world.remove(world.get(PointerSystem))), nothing flushes those arrays, so they grow without bound on every native pointer event — a memory leak (as noted by @eonarheim, pointer flushing should not depend on the system being present).

• src/engine/input/input-host.ts: InputHost.update() (which runs every frame for both the engine-level and scene-level input hosts) now guarantees the pointer receiver is flushed exactly once per frame, independent of the PointerSystem. If the receiver was already processed by the system this frame, it just resets the marker; otherwise it calls update() + clear() itself.
• src/engine/input/pointer-system.ts: after the system updates/clears each receiver it marks receiver._processedThisFrame = true so the InputHost fallback does not double-process (and therefore never double-emits raw pointer events) when a PointerSystem is active.
• src/engine/input/pointer-event-receiver.ts: added the internal _processedThisFrame flag used to coordinate the two.
• src/spec/vitest/pointer-input-spec.ts: added a regression test that removes the PointerSystem from the scene, dispatches several pointer events across multiple frames, and asserts the receiver's currentFrame* arrays stay empty.
• CHANGELOG.md: added a Fixed entry.

Note: the project's browser-mode (Playwright/Chromium) test runner could not launch a browser in my local sandbox (browserType.launch: spawn UNKNOWN), so the suite was validated via tsc on the engine sources; the new spec follows the existing patterns in the same file and should run normally in CI.