fix(models): standardize Anthropic API error handling in AnthropicLlm

[sebastienc]

Summary

Fixes #5396

Disclaimer: I'm using Claude Code to generate the following text. I find the summary is better than what I would've wrote.

Standardizes how AnthropicLlm handles errors from the Anthropic API, bringing it in line with the error handling pattern used elsewhere in ADK.

• AnthropicRateLimitError — new exception class (subclasses Exception, not anthropic.RateLimitError) raised when the API returns a 429; includes a helpful mitigation message pointing users toward rate limit remediation options; original error is chained via from e so the full traceback is preserved
• generate_content_async — wraps the non-streaming messages.create call in a try/except anthropic.RateLimitError block and re-raises as AnthropicRateLimitError
• _generate_content_streaming — same wrapping around the streaming messages.create call; all other anthropic exceptions propagate unchanged

Testing plan

4 new tests added to tests/unittests/models/test_anthropic_llm.py. Full suite: 63 passed.

tests/unittests/models/test_anthropic_llm.py::test_non_streaming_rate_limit_raises_anthropic_rate_limit_error PASSED
tests/unittests/models/test_anthropic_llm.py::test_streaming_rate_limit_raises_anthropic_rate_limit_error PASSED
tests/unittests/models/test_anthropic_llm.py::test_non_streaming_other_errors_propagate PASSED
tests/unittests/models/test_anthropic_llm.py::test_streaming_other_errors_propagate PASSED