feat: use API-based token counting for 100% accuracy

Switch from local estimation (gpt-tokenizer) to API-based counting
via Google Cloud Code API for accurate token counts. Falls back to
local estimation if API call fails.

src/cloudcode/count-tokens.js

@@ -278,7 +278,7 @@ export function createCountTokensHandler(accountManager) {
             const result = await countTokens(
                 { messages, model, system, tools, tool_choice, thinking },
                 accountManager,
-                { useAPI: false } // Default to local estimation for speed
+                { useAPI: true } // Use API for accurate token counting
             );
 
             res.json(result);