Simple RNN Cell

Implement one step of a vanilla RNN cell.

The RNN update equations: $$h_t = \tanh(x_t \cdot W_{xh} + h_{t-1} \cdot W_{hh} + b_h)$$

Process a sequence of inputs and return all hidden states.

Input:

• x: input sequence of shape (seq_len, input_dim)
• W_xh: input-to-hidden weights of shape (input_dim, hidden_dim)
• W_hh: hidden-to-hidden weights of shape (hidden_dim, hidden_dim)
• b_h: hidden bias of shape (hidden_dim,)
• h0: initial hidden state of shape (hidden_dim,)

Output: All hidden states, shape (seq_len, hidden_dim).