test: stabilize UDP mux close assertion

[mertushka]

Summary

Stabilize the UDP mux listener close-delivery regression test for the data-channel-first nonstandard UDP mux path.

The test still verifies that close() / stop() suppress callbacks after close, but it no longer assumes the pre-close STUN send loop delivers exactly one callback before the close point. On Windows, the existing sendUntil loop can observe more than one valid pre-close callback before listener.close() runs.

Verification

• npm run check
• npm run native:check - not applicable; no native source change
• npm run build - not applicable; test-only change
• npm test
• npm run api:check - not applicable; no API surface change
• npm run types:check - not applicable; no type change
• npm run e2e:chrome - not applicable; no browser interop change
• npm run wpt:selection:check - not applicable; no WPT selection change
• npm run wpt:smoke - not applicable; no WPT harness or WebRTC facade change
• npm run wpt:smoke:check - not applicable; no WPT harness or WebRTC facade change
• npm run wpt:test / npm run wpt:check:strict - not run locally; this fixes a pre-WPT local test failure in Conformance

Also run locally: node --test --test-timeout=30000 test/nonstandard.test.js, git diff --check.

WPT impact

No wpt-manifest.json, selected WPT result, or docs/divergences.md change.

Browser interoperability

No Chrome E2E behavior or coverage change.

Notes

This addresses the manual Conformance failure on windows-latest / Node 20 / WPT, where the pre-WPT npm test step failed because the test asserted total callback count instead of post-close callback stability.