Recurrent Depth Model Implementation

This module implements the recurrent depth transformer model as described in
'Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach'.

The model consists of three main components:

• Prelude: Embeds input tokens into latent space
• Core/Recurrent: Iteratively processes the latent state
• Coda: Decodes the final latent state into output probabilities