Add rate limit retry logic to upgrade-ai command

[zhangzhenghao]

Add exponential backoff retry logic for OpenAI API 429 (Too Many Requests) errors in the upgrade-ai command.

Changes:

• Import RateLimitError from openai
• Add call_isai_with_retry() function with exponential backoff (default: 5 retries, base delay 1s)
• Update upgrade_ai() to use retry wrapper
• Add explicit handling for RateLimitError after retries exhausted

The retry logic:

• Initial delay: 1 second
• Each retry doubles the delay (1s → 2s → 4s → 8s → 16s)
• Max 5 retries before giving up
• Other exceptions are passed through without retry