fix(azure): nest <prosody> inside <voice>, not as direct child of <speak>

[OwenMcGirr]

Problem

When rate, pitch, or volume are passed as options to synthToBytestream / synthToBytes, the generated SSML places <prosody> as a direct child of <speak> instead of inside <voice>:

<!-- Generated (invalid) -->
<speak ...>
  <prosody rate="fast" pitch="high" volume="80%">
    <voice name="en-US-JennyNeural">Hello world</voice>
  </prosody>
</speak>

Azure rejects this with HTTP 500:

Node [speak] with type [RootSpeak] should not contain node [prosody] with type [Others].

Root cause

In ensureAzureSSMLStructure, the this.properties branch (rate/pitch/volume set on the client) correctly extracts content from inside <voice>. But the options branch (rate/pitch/volume passed per-call) always extracted from inside <speak>, so once <voice> had been injected earlier in the same method, the prosody ended up wrapping it from the outside.

Fix

Mirror the this.properties branch in the options branch: when <voice> is present, replace content inside <voice>; only fall back to <speak> when no <voice> element exists.

<!-- Generated after fix (valid) -->
<speak ...>
  <voice name="en-US-JennyNeural">
    <prosody rate="fast" pitch="high" volume="80%">Hello world</prosody>
  </voice>
</speak>

Test

Added regression test in azure-mstts-namespace.test.ts that asserts <prosody> appears after <voice> in the output and matches /<voice[^>]*>\s*<prosody[^>]*>/.

Closes #38