Fix missing error handling in LLM response stream iteration

[Copilot]

The summarizer was iterating over the LLM response stream without checking errors, causing API failures to silently result in nil content.

Changes:

• Check err in response stream iteration (line 62-65)
• Fix Parts type from []genai.Part to []*genai.Part for genai API compatibility
• Fix Part initialization from genai.Text(prompt) to {Text: prompt}

Before:

for resp := range responseStream {
    if resp == nil {
        continue
    }
    if resp.Content != nil {
        summaryContent = resp.Content
        break
    }
}

After:

for resp, err := range responseStream {
    if err != nil {
        return nil, fmt.Errorf("LLM generation error: %w", err)
    }
    if resp == nil {
        continue
    }
    if resp.Content != nil {
        summaryContent = resp.Content
        break
    }
}

LLM API failures now surface immediately with context instead of returning nil summary content.

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.