Fix sliding window context management handling of images and get more accurate token estimates

package.json

@@ -136,13 +136,16 @@
     "@anthropic-ai/sdk": "^0.26.0",
     "@anthropic-ai/tokenizer": "^0.0.4",
     "@kodu-ai/cloud-api": "^1.0.1",
+    "@types/clone-deep": "^4.0.4",
     "@vscode/codicons": "^0.0.36",
     "axios": "^1.7.4",
+    "clone-deep": "^4.0.1",
     "default-shell": "^2.2.0",
     "delay": "^6.0.0",
     "diff": "^5.2.0",
     "execa": "^9.3.0",
     "globby": "^14.0.2",
+    "image-size": "^1.1.1",
     "openai": "^4.54.0",
     "os-name": "^6.0.0",
     "p-wait-for": "^5.0.2",