add optional checkpoint saving to Trainer Class

[spencerirving]

Description

Adds optional checkpoint saving functionality to the Trainer class, enabling users to save model state during training without modifying existing behavior.

This fixes issue #48

Changes Made

Added checkpoint parameters to Trainer.__init__():

• checkpoint_dir: Directory for saving checkpoints
• checkpoint_epochs: When to save(int for intervals, list for specific epochs)
• checkpoint_name: Base filename for checkpoints
• save_best: Whether to track and save best model

Added private helper methods:

• _should_save_checkpoint(): Determines when to save based setup
• _save_checkpoint(): Handles checkpoint file creation

Added public method:

• load_checkpoint: Restores model and optimizer state from checkpoint

Integrated checkpoint logic into 'fit()' method.

Usage

Basic Usage - Save Every 10 Epochs

trainer = Trainer(
model, loss_fn, optimizer, 
epochs=100, 
checkpoint_dir="./checkpoints", checkpoint_epochs=10, 
checkpoint_name="my_model"
)
trainer.fit(train_loader)

Track Best Model

trainer = Trainer(
model, loss_fn, optimizer, 
epochs=100, 
checkpoint_dir="./checkpoints", checkpoint_epochs=10, 
checkpoint_name="my_model",
save_best=True # Saves best model as my_model_best.pt
)
trainer.fit(train_loader)