Can large language models estimate public opinion about global warming? An empirical assessment of algorithmic fidelity and bias