test(voice): use AudioFrame.timestamp (not timestampMs) in STT regression test

CI typecheck failed because AudioFrame.timestamp is the canonical field;
timestampMs was a slip in the regression test I just added. Spec still
passes — only the property name was wrong.

Author: jddunn

src/io/voice-pipeline/__tests__/provider-health.spec.ts

@@ -107,7 +107,7 @@ describe('provider HealthyProvider implementations', () => {
       session.pushAudio({
         samples,
         sampleRate: 16000,
-        timestampMs: 0,
+        timestamp: 0,
       });
       await session.flush();