[Feature] : Implement Zero-Knowledge Proof of Dataset Inclusion (ZK-DI)

### What is ZK-DI?
ZK-DI allows a prover to cryptographically prove to a verifier that a specific chunk of text *is exactly present* in the dataset used for training, **without** the verifier needing to download the entire dataset. The verifier only needs the global `merkle_root` and the ZK Proof.

ZK-DI is entirely independent of the neural network or the model training code. ZK-DI operates strictly on the **Data Layer** (which we just finished building) and the **Cryptographic Prover Layer**