test(node): stabilize pubsub channels (unique runId + bounded wait)

[hank95179]

Issue link

This Pull Request is linked to issue (URL): #4923

What & Why (Summary)

The PubSub › test pubsub channels_%p test sporadically misses test_channel2. Root causes are cross-run/parallel contamination (shared channel names) and a small timing gap between subscription establishment and observability via pubsubChannels().

This PR stabilizes the test with minimal changes:

1. Unique naming isolation: Introduce a per-run runId and suffix all channels/patterns with :${runId} to avoid interference across tests or previous runs.
2. Bounded wait before assert: Add a short polling window (up to 60 iterations × 50 ms) before final assertions so the test observes the expected set once subscriptions are visible.

No production code is changed; only the test data and assert timing are adjusted.

Details of the change

• Add const runId = String(getRandomKey());
• Change test_channel1|test_channel2|some_channel3 and patterns to …:${runId}
• Poll pubsubChannels() (and pubsubChannels({ pattern })) until the exact expected set is observed or the bounded window expires, then perform the final expect.
• Bind the “non-matching pattern” to non_matching_*:${runId} to avoid accidental matches from other runs.

Why this fixes the flake

• Isolation removes cross-test contamination from shared channel names.
• Bounded wait absorbs the small timing window between subscription setup and visibility in pubsubChannels().

Risk & Alternatives

• Low risk: Test-only change; does not affect runtime behavior.
• Alternatives (e.g., enforcing global serial execution or heavy global cleanup) are broader in scope and slow down CI. This PR keeps the fix local and minimal.

How I verified

• Repeated runs in local/container environments for tests/PubSub.test.ts targeting channels_true (both in-band and parallel CI-style) no longer show the missing test_channel2 diff.
• Other PubSub tests remain unaffected.