feat: implement fuzzy search and replace functionality with Levenshtein distance

- Added `applySearchReplace` function to handle search and replace operations with fuzzy matching capabilities.
- Introduced tests for various scenarios including fuzzy matching with typos, exact matches, and handling whitespace differences.
- Created a parser for search/replace blocks to facilitate the new functionality.
- Updated prompts for search-replace operations to clarify usage and examples.
- Added utility functions for text normalization and language detection based on file extensions.
- Implemented a minimal stdio MCP server for local testing with tools for adding numbers and printing environment variables.

testing/fake-llm-server/index.ts

@@ -137,13 +137,20 @@ app.get("/lmstudio/api/v0/models", (req, res) => {
   res.json(lmStudioModels);
 });
 
-["lmstudio", "gateway", "engine", "ollama"].forEach((provider) => {
+["lmstudio", "gateway", "engine", "ollama", "azure"].forEach((provider) => {
   app.post(
     `/${provider}/v1/chat/completions`,
     createChatCompletionHandler(provider),
   );
 });
 
+// Azure-specific endpoints (Azure client uses different URL patterns)
+app.post("/azure/chat/completions", createChatCompletionHandler("azure"));
+app.post(
+  "/azure/openai/deployments/:deploymentId/chat/completions",
+  createChatCompletionHandler("azure"),
+);
+
 // Default test provider handler:
 app.post("/v1/chat/completions", createChatCompletionHandler("."));