Do LLMs performing chain-of-thought reasoning implement genuine neurosymbolic processing with symbol-like internal representations?