refactor: remove unused qwen-integration feature and dead code

Architectural cleanup based on codebase analysis that revealed qwen-integration
feature flag gates completely unused code. Qwen models work perfectly via
Ollama and LM Studio providers without any special compilation flags.

Removed:
- qwen-integration feature flag from Cargo.toml
- 7 unused module files (~3,800 LOC):
  - cache.rs (18K) - Qwen response caching
  - context_optimizer.rs (11K) - Qwen context optimization
  - performance.rs (7.4K) - Qwen performance monitoring
  - pattern_detector.rs (23K) - Pattern detection
  - qwen.rs (8.7K) - QwenClient implementation
  - qwen_prompts.rs (12K) - Qwen-specific prompts
  - tools_schema.rs - Qwen tool schemas
- Module gates from lib.rs
- QwenClient field and methods from official_server.rs:
  - qwen_client field declaration and initialization
  - initialize_qwen() method (never called)
  - get_qwen_client() method (never called)
- initialize_qwen() call from bin/codegraph.rs

Updated:
- install-codegraph.sh - removed qwen-integration from FEATURES
- install-codegraph-local-speed-osx.sh - removed qwen-integration from FEATURES

Why This Is Safe:
1. None of the gated modules were actually instantiated or used
2. get_qwen_client() method was defined but never called
3. Qwen models work via existing Ollama/LM Studio providers:
   - CODEGRAPH_LLM_PROVIDER=ollama (runs Qwen via Ollama)
   - CODEGRAPH_LLM_PROVIDER=lmstudio (runs Qwen via LM Studio)
4. The qwen_simple.rs provider in codegraph-ai is always available
5. Build succeeds with all tests passing

Impact:
- ~3,800 lines of dead code removed
- Simplified architecture (one less feature flag to maintain)
- No functional changes - Qwen models still work
- Cleaner codebase with less technical debt

The qwen-integration flag was a relic of an earlier architecture where
Qwen had special first-class integration. Modern CodeGraph uses generic
LLM providers, making this special-casing unnecessary.

Author: jruokola

crates/codegraph-mcp/Cargo.toml

@@ -91,6 +91,5 @@ embeddings-jina = ["embeddings", "codegraph-vector/jina"]
 embeddings-lmstudio = ["embeddings", "codegraph-vector/lmstudio"]
 cloud = ["embeddings-jina", "codegraph-graph/surrealdb"]
 server-http = ["dep:axum", "dep:hyper", "dep:tower", "dep:http-body-util"]
-qwen-integration = []
 ai-enhanced = ["dep:codegraph-ai", "embeddings", "autoagents-experimental", "codegraph-ai/openai-compatible"]
 autoagents-experimental = ["dep:autoagents", "dep:autoagents-derive"]

crates/codegraph-mcp/src/bin/codegraph.rs

@@ -892,7 +892,6 @@ async fn handle_start(
 
             // Create and initialize the revolutionary CodeGraph server with official SDK
             let server = codegraph_mcp::official_server::CodeGraphMCPServer::new();
-            server.initialize_qwen().await;
 
             if atty::is(Stream::Stderr) {
                 eprintln!(