feat(voice): add first-class voice support primitives

[drewdrewthis]

• Add AudioData type and utilities (audioFromFile, audioFromBase64, audioFromBuffer)
• Add textToSpeech utility for TTS conversion
• Add scenario.user.speak() and scenario.agent.speak() for TTS messages
• Add voice option to userSimulatorAgent for audio output
• Add audio option to judgeAgent for multimodal evaluation

Proposed API:

    const result = await scenario.run({
      name: "fixed voice messages",
      description: "Test with predetermined voice messages",
      agents: [
        new VoiceAgent(),
        scenario.userSimulatorAgent({
          voice: "nova",
        }), // Text sim (not used in this script)
        scenario.judgeAgent({
          criteria: ["Agent responds appropriately to greeting"],
          // audio: "transcribe" | true | undefined,
        }),
      ],
      script: [
        // Fixed user voice message via TTS
        scenario.user.speak("Hello! Can you help me with something?"),
        scenario.agent(), // Agent generates audio response
        async (ctx) => {
          await saveConversationAudio(
            ctx,
            path.join(
              outputPath,
              `${StringUtils.kebabCase(ctx.config.name)}.wav`
            )
          );
        },
        scenario.judge(),
      ],
      setId,
    });
    ```

[drewdrewthis]

Suggested change

	audio: true,
	transcribeOnly?: true

[drewdrewthis]

Suggested change

	* user("Help me with billing"), // Fixed text
	* user.text("Help me with billing"), // Fixed text

otherwise it uses its default?