fix: voice selection on iOS/Mac (Web Speech API)

[klokie]

Fixes #42, fixes #20.

Problem

Two bugs prevent the selected voice from being used on iOS and macOS:

Bug 1: Voice matched by name but stored as voiceURI

getVoices() stores voice identity as voiceURI, which becomes the saved setting value. But sayWithVoice looked up the voice by name:

// before
msg.voice = window.speechSynthesis.getVoices()
    .filter(otherVoice => otherVoice.name === voice)[0];

On desktop (Chrome/Firefox), voiceURI and name are often identical so this accidentally worked. On iOS/macOS they differ — Siri and enhanced voices have URIs like com.apple.voice.compact.en-US.Samantha vs display name "Samantha (Enhanced)" — so the filter returns undefined and playback falls back to the system default every time.

Bug 2: Synchronous getVoices() on iOS/Safari

window.speechSynthesis.getVoices() returns an empty array synchronously on iOS/Safari. Voices load asynchronously and are only available after the voiceschanged event fires. Both getVoices() and sayWithVoice called it synchronously, so on iOS the settings dropdown could appear empty and msg.voice was always undefined.

Fix

• Match voices by voiceURI instead of name in sayWithVoice
• Add a loadVoices() helper that returns voices immediately if already available, otherwise awaits voiceschanged
• Use loadVoices() in both getVoices() and sayWithVoice