fix(api): update server busy message

[shihaobai]

Summary

• change the default server-busy message to Too many requests. Please try again later.
• return the exception message without appending the status code
• update OpenAI-compatible and Anthropic regression expectations

Why

The existing 429 handling still returns the old Server is busy text and appends (Status code: 429) through ServerBusyError.__str__. This makes the response message differ from the intended rate-limit message even though the HTTP status is correct.

Impact

Server-busy responses continue to use HTTP 429 and the existing rate-limit error types. Their message is now exactly Too many requests. Please try again later. across response envelopes.

Validation

• python3 -m py_compile lightllm/utils/error_utils.py test/test_api/test_server_busy_handling.py
• isolated assertions for the default status, message, and string representation
• git diff --check

The full pytest target was not run locally because the LightLLM runtime dependencies, including pytest and torch, are not installed in this environment.