Text this: VAE+DDPG: An Attention‐Enhanced Variational Autoencoder for Deep Reinforcement Learning‐Based Autonomous Navigation in Low‐Light Environments