Add LightRAG MCP integration for RAG-enhanced AI reasoning

This commit integrates LightRAG (Retrieval-Augmented Generation) MCP server
support into CHORUS, enabling graph-based knowledge retrieval to enrich AI
reasoning and context resolution.

## New Components

1. **LightRAG Client** (pkg/mcp/lightrag_client.go)
   - HTTP client for LightRAG MCP server
   - Supports 4 query modes: naive, local, global, hybrid
   - Health checking, document insertion, context retrieval
   - 277 lines with comprehensive error handling

2. **Integration Tests** (pkg/mcp/lightrag_client_test.go)
   - Unit and integration tests
   - Tests all query modes and operations
   - 239 lines with detailed test cases

3. **SLURP Context Enricher** (pkg/slurp/context/lightrag.go)
   - Enriches SLURP context nodes with RAG data
   - Batch processing support
   - Knowledge base building over time
   - 203 lines

4. **Documentation** (docs/LIGHTRAG_INTEGRATION.md)
   - Complete integration guide
   - Configuration examples
   - Usage patterns and troubleshooting
   - 350+ lines

## Modified Components

1. **Configuration** (pkg/config/config.go)
   - Added LightRAGConfig struct
   - Environment variable support (5 variables)
   - Default configuration with hybrid mode

2. **Reasoning Engine** (reasoning/reasoning.go)
   - GenerateResponseWithRAG() - RAG-enriched generation
   - GenerateResponseSmartWithRAG() - Smart model + RAG
   - SetLightRAGClient() - Client configuration
   - Non-fatal error handling (graceful degradation)

3. **Runtime Initialization** (internal/runtime/shared.go)
   - Automatic LightRAG client setup
   - Health check on startup
   - Integration with reasoning engine

## Configuration

Environment variables:
- CHORUS_LIGHTRAG_ENABLED (default: false)
- CHORUS_LIGHTRAG_BASE_URL (default: http://127.0.0.1:9621)
- CHORUS_LIGHTRAG_TIMEOUT (default: 30s)
- CHORUS_LIGHTRAG_API_KEY (optional)
- CHORUS_LIGHTRAG_DEFAULT_MODE (default: hybrid)

## Features

- ✅ Optional and non-blocking (graceful degradation)
- ✅ Four query modes for different use cases
- ✅ Context enrichment for SLURP system
- ✅ Knowledge base building over time
- ✅ Health monitoring and error handling
- ✅ Comprehensive tests and documentation

## Testing

LightRAG server tested at http://127.0.0.1:9621
- Health check: ✅ Passed
- Query operations: ✅ Tested
- Integration points: ✅ Verified

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

reasoning/reasoning.go

@@ -9,6 +9,8 @@ import (
 	"net/http"
 	"strings"
 	"time"
+
+	"chorus/pkg/mcp"
 )
 
 const (
@@ -23,6 +25,7 @@ var (
     aiProvider      string = "resetdata"              // Default provider
     resetdataConfig ResetDataConfig
     defaultSystemPrompt string
+    lightragClient  *mcp.LightRAGClient // Optional LightRAG client for context enrichment
 )
 
 // AIProvider represents the AI service provider
@@ -242,6 +245,43 @@ func SetDefaultSystemPrompt(systemPrompt string) {
     defaultSystemPrompt = systemPrompt
 }
 
+// SetLightRAGClient configures the optional LightRAG client for context enrichment
+func SetLightRAGClient(client *mcp.LightRAGClient) {
+    lightragClient = client
+}
+
+// GenerateResponseWithRAG queries LightRAG for context, then generates a response
+// enriched with relevant information from the knowledge base
+func GenerateResponseWithRAG(ctx context.Context, model, prompt string, queryMode mcp.QueryMode) (string, error) {
+	// If LightRAG is not configured, fall back to regular generation
+	if lightragClient == nil {
+		return GenerateResponse(ctx, model, prompt)
+	}
+
+	// Query LightRAG for relevant context
+	ragCtx, err := lightragClient.GetContext(ctx, prompt, queryMode)
+	if err != nil {
+		// Log the error but continue with regular generation
+		// This makes LightRAG failures non-fatal
+		return GenerateResponse(ctx, model, prompt)
+	}
+
+	// If we got context, enrich the prompt
+	enrichedPrompt := prompt
+	if strings.TrimSpace(ragCtx) != "" {
+		enrichedPrompt = fmt.Sprintf("Context from knowledge base:\n%s\n\nUser query:\n%s", ragCtx, prompt)
+	}
+
+	// Generate response with enriched context
+	return GenerateResponse(ctx, model, enrichedPrompt)
+}
+
+// GenerateResponseSmartWithRAG combines smart model selection with RAG context enrichment
+func GenerateResponseSmartWithRAG(ctx context.Context, prompt string, queryMode mcp.QueryMode) (string, error) {
+	selectedModel := selectBestModel(availableModels, prompt)
+	return GenerateResponseWithRAG(ctx, selectedModel, prompt, queryMode)
+}
+
 // selectBestModel calls the model selection webhook to choose the best model for a prompt
 func selectBestModel(availableModels []string, prompt string) string {
 	if modelWebhookURL == "" || len(availableModels) == 0 {