fix(anthropic): warn when stop_reason='max_tokens' silently truncates output

[Beandon13]

Summary

• AnthropicCompletion previously discarded response.stop_reason after extracting token usage, so callers had no way to detect that a Claude response was cut off by max_tokens. The follow-up call inside _handle_tool_use_conversation is the most painful site because the truncated synthesis answer is what downstream code parses.
• This PR adds a small module-level helper, _warn_on_max_tokens_truncation, and calls it at all four completion sites (sync/async × plain completion / tool-use follow-up). When stop_reason == "max_tokens" we emit logging.warning(...) including the agent role (when available) and the configured max_tokens so users can act on it.
• Behavior is otherwise unchanged. No new public API; no extra LLM calls; no events touched.

Closes #5148

Testing

cd lib/crewai
uv pip install -e ".[anthropic]"
uv pip install pytest pytest-asyncio pytest-xdist pytest-timeout pytest-recording vcrpy
python -m pytest \
  tests/llms/anthropic/test_anthropic.py::test_warn_on_max_tokens_truncation_emits_warning \
  tests/llms/anthropic/test_anthropic.py::test_warn_on_max_tokens_truncation_silent_for_normal_stop \
  tests/llms/anthropic/test_anthropic.py::test_warn_on_max_tokens_truncation_includes_agent_role \
  tests/llms/anthropic/test_anthropic.py::test_handle_completion_logs_truncation_warning

``` ============================== 4 passed in 7.31s =============================== ```

Existing token-usage tests in the same file (test_anthropic_token_usage_tracking, test_anthropic_cache_creation_tokens_extraction, test_anthropic_missing_cache_fields_default_to_zero, test_anthropic_completion_is_used_when_anthropic_provider) also pass unchanged.