feat: switch to Qwen3-0.6B for summarization

[joemckenney]

Summary

• Upgrade LLM from Qwen2.5-0.5B to Qwen3-0.6B (397MB vs 491MB - 19% smaller)
• Simplify summarization prompt to just "Summarize this terminal output in 1-2 sentences."
• Add /no_think directive to disable Qwen3's reasoning mode for faster, more concise output
• Add strip_think_blocks() fallback to filter any <think> tags from output

Why Qwen3?

• Newer model with improved instruction-following
• Smaller download size despite more parameters
• Better summarization quality with simpler prompts
• Official GGUF support via unsloth/Qwen3-0.6B-GGUF

Test plan

• cargo test passes
• Manual testing: summaries are concise and accurate
• Model downloads successfully from HuggingFace
• /no_think prevents verbose reasoning output