feat(training): add validation split, early stopping, checkpointing, …

[SUJALGOYALL]

Description

This pull request improves the LSTM model training loop by adding several enhancements to improve generalization, prevent overfitting, and stabilize learning.

Key Updates

• Added a validation split (~15%) for monitoring validation metrics.
• Implemented early stopping with best-weight restoration.
• Added model checkpointing to automatically save the best model.
• Applied ReduceLROnPlateau scheduler to dynamically adjust the learning rate when validation loss stalls.

These changes make the training process more efficient and robust while reducing overfitting.

---

Semver Changes

• Patch (bug fix, no new features)
• Minor (new features, no breaking changes)
• Major (breaking changes)

---

Issues

Closes #3 >

---

Checklist

• I have read the Contributing Guidelines