L-Dynamic Attention: Learned Age-Aware KV Cache Management for Efficient Long-Context Transformers | Synapse