feat: Add token rate limiter to prevent Anthropic/OpenAI rate limit errors

[im0xMagnus]

Rate Limiter for Anthropic API Requests

Problem

OpenManus users are experiencing rate limit errors (HTTP 429) when using Anthropic's Claude 3.7 API, which has a limit of 20,000 input tokens per minute. The current retry mechanism doesn't effectively handle these rate limits.

Solution

This PR adds:

1. A token rate limiter that tracks API usage and proactively waits when approaching limits
2. A more intelligent retry strategy specifically for rate limit errors
3. Proper Anthropic API integration with message format conversion
4. Token estimation to predict usage before sending requests

Benefits

• Prevents API rate limit errors by managing token usage proactively
• Provides informative log messages when approaching limits
• Makes OpenManus more resilient during heavy usage
• Improves user experience by reducing errors

Testing

Tested with Anthropic APIs under heavy load scenarios.

[hyperchi]

this is what i was trying to solve too! TY so much.

BTW, can you also add readme and associated config exmaple changes for ppl to read

[im0xMagnus]

this is what i was trying to solve too! TY so much.

BTW, can you also add readme and associated config exmaple changes for ppl to read

Will work on that next for you