[#9] feat-003 Add ReasoningState Before Planning and UI updates

[cymtrick]

Add ReasoningState with LLM Analysis Before Planning

🎯 Problem

Saturn was jumping directly from StartState to PlanningState without any preliminary analysis of user queries. This resulted in:

• No visual feedback during the initial thinking process
• LLM planning without structured reasoning about query complexity and requirements
• Suboptimal execution plans due to lack of upfront analysis

🚀 Solution

Added a new ReasoningState that performs actual LLM-based analysis before planning, with dynamic visual feedback showing the thinking process.

✨ What's New

New State Flow

Before: StartState → PlanningState → ExecutingState
After: StartState → ReasoningState → PlanningState → ExecutingState

Chat Interface (saturn chat):

• Same 3-line reasoning phases
• Proper cleanup of thinking messages
• Smooth transition to normal assistant responses

🧪 Testing

Manual Testing

# Test in CLI
saturn run "Create a complex multi-region GCP infrastructure"

# Test in chat  
saturn chat
> Deploy a machine learning pipeline with security scanning

Expected Behavior

1. Shows dynamic 3-line thinking display during reasoning
2. LLM performs actual analysis of the query
3. Planning prompt includes reasoning context
4. Better execution plans based on upfront analysis

📊 Performance Impact

• Additional LLM Call: One reasoning call before planning (~1-2 seconds)
• Improved Planning: Better execution plans due to structured analysis
• User Experience: Clear visual feedback that Saturn is actively thinking

🔄 Backwards Compatibility

• ✅ All existing commands work unchanged
• ✅ No breaking changes to APIs or interfaces
• ✅ Graceful fallback if reasoning LLM call fails

🎨 UI/UX Improvements

• Visual Feedback: Users see Saturn "thinking" with detailed phase descriptions
• Better Context: Planning benefits from structured reasoning analysis
• Professional Feel: Multi-line animations similar to modern AI interfaces

📝 Configuration

No new configuration required - works with existing LLM provider settings.

🚨 Error Handling

• Graceful fallback if LLM reasoning fails
• Visual display cancellation on errors
• Structured error logging in state recorder

🔮 Future Enhancements

• Reasoning complexity could influence parallel execution strategy
• Domain-specific reasoning prompts (ML, security, infrastructure)
• Reasoning cache for similar queries
• User-configurable reasoning depth