feat: add real-time streaming for generator tool preliminary results

[mattapperson]

Summary

• Adds real-time streaming of generator tool preliminary results as they yield, rather than batch-emitting after tool execution completes
• Introduces ToolEventBroadcaster class for push-based multi-consumer event delivery following the ReusableReadableStream pattern
• Updates getToolStream() and getFullResponsesStream() to deliver events in real-time

Motivation

Similar to Vercel AI SDK v6's streaming preliminary tool outputs feature. This enables streaming UI updates during long-running tool operations - users can see progress as it happens rather than waiting for completion.

Changes

• New: src/lib/tool-event-broadcaster.ts - Push-based broadcaster with buffering and multiple consumer support
• Modified: src/lib/model-result.ts - Added executeToolsWithBroadcast() method, updated stream methods
• Export: ToolEventBroadcaster exported for advanced use cases

Usage

const progressTool = tool({
  name: 'progress_task',
  inputSchema: z.object({ task: z.string() }),
  eventSchema: z.object({ progress: z.number() }),
  outputSchema: z.object({ completed: z.boolean() }),
  execute: async function* (params) {
    yield { progress: 25 };  // Streamed immediately!
    await doWork();
    yield { progress: 50 };  // Streamed immediately!
    yield { completed: true };
  },
});

// Consume real-time events
for await (const event of response.getToolStream()) {
  if (event.type === 'preliminary_result') {
    console.log(`Progress: ${event.result.progress}%`);
  }
}

Test plan

• Unit tests for ToolEventBroadcaster (11 tests covering single/multiple consumers, async waiting, error handling)
• E2E tests for real-time streaming via getToolStream() and getFullResponsesStream()
• All existing tests pass