Fix and update the token limits handling

[zhutao100]

This PR enhances the token limits functionality by:

1. Adding support for new models:

   • GLM-4.6 with 200k token limit
   • DeepSeek-v3.2-exp with 128k token limit

2. Improving the model name normalization logic to:

   • Handle more date-like suffixes (e.g., deepseek-r1-0528)
   • Handle -exp suffixes in model names
   • Better distinguish between version numbers that should be stripped vs. kept
   • Improve handling of version numbers with and without the 'v' prefix

The changes are focused on improving the accuracy of token limit detection for various AI models, which is critical for ensuring proper handling of different model capabilities in the Qwen Code CLI.

Testing Matrix

	🍏	🪟	🐧
npm run	👍	❓	❓
npx	👍	❓	❓
Docker	❓	❓	❓
Podman	❓	-	-
Seatbelt	❓	-	-

[Mingholy]

LGTM