An RNN forward pass

Implement embedding lookup and a vanilla RNN over a whole sequence.

• embed(tokens, E): given integer token ids (a 1-D array) and an embedding matrix E (V, D), return the (T, D) array of their vectors.
• rnn_forward(X, h0, Wx, Wh, b): X has shape (T, D), h0 shape (H,), Wx (D, H), Wh (H, H), b (H,). Apply ${\mathbf{h}}_t=\tanh ({\mathbf{x}}_t{W}_x+{\mathbf{h}}_{t-1}{W}_h+\mathbf{b})$ for each step and return all hidden states stacked, shape (T, H).