(issue #67) make early_stopping optional; cpu default device for trainer, evaluator

[ChrisNick92]

Make Early Stopping Optional & Add current_patience to Trainer State

Overview

Early stopping is now opt-in rather than always active, current_patience is exposed on trainer.state for runtime introspection, and the default device for Trainer and Evaluator is changed to "cpu" to avoid errors on machines without a GPU.

Changes

Early Stopping — opt-in

• patience now defaults to None (previously 15). When None, EarlyStopper is not added to the callback list and training always runs for the full number of epochs.
• EarlyStopper is only instantiated when patience is explicitly set.
• Callback list typed as list[BaseCallback] to correctly accommodate both Checkpointer and EarlyStopper.

current_patience on State

• State.current_patience: int = 0 added — updated by EarlyStopper after each epoch, allowing callers to monitor patience consumption via trainer.state.current_patience.

Default device changed to "cpu"

• Trainer, Evaluator, and get_device now default to device="cpu". Previously defaulting to "cuda" caused an immediate ValueError on CPU-only machines before any training could start.
• get_device error messages updated to suggest the correct alternative device.