fix(anthropic): preserve thinking blocks in output spans

[chocoHacks33]

Summary

• map Anthropic thinking / beta thinking blocks to Logfire semconv reasoning parts
• preserve Anthropic thinking content in legacy response_data.message for the default instrumentation path
• add regression coverage for both the semconv conversion helper and legacy response payload

Why

Fixes the missing Anthropic thinking-block capture behind issue #1737, where extended/interleaved thinking content was not preserved cleanly in Logfire's output span data.

Validation

• direct Python smoke check for convert_response_to_semconv(...) and on_response(..., version=1) with ThinkingBlock
• git diff --check

Environment limits

• pytest tests/otel_integrations/test_anthropic.py ... is currently blocked in this environment by an import-time PendingDeprecationWarning from the MCP test bootstrap (python_multipart / multipart)
• python -m compileall ... hit local OSError: [Errno 28] No space left on device