Module trainer 

Training loop with progress monitoring, learning rate decay, and optional checkpointing.

Training Flow

corpus ──► Vocabulary::build ──► Model::new
                                      │
             ┌────────────────────────┘
             ▼
        for each epoch:
          shuffle sentences
          for each sentence:
            subsample tokens
            for each (center, context) pair:
              sample negatives
              Model::update (SGD step)
              update LR (linear decay)
          record epoch loss
             │
             ▼
        Embeddings ──► .most_similar() / .analogy() / .save()

Structs

EpochStats

Recorded statistics for one epoch.

Trainer

Manages the full training pipeline.