Attentive Temporal Pooling for Conformer-based Streaming Language Identification in Long-form Speech