Text this: Adaptive Channel-Enhanced Graph Convolution for Skeleton-Based Human Action Recognition