feat(handlers): rename EarlyStopping params for consistency with PyTorch

[joemunene-by]

Summary

Addresses the remaining work in #3411 — renames EarlyStopping parameters to align with PyTorch's ReduceLROnPlateau naming conventions, with full backward compatibility and deprecation warnings.

Parameter Renames

Old Name	New Name	Rationale
min_delta	threshold	Matches ReduceLROnPlateau.threshold
min_delta_mode	threshold_mode	Matches ReduceLROnPlateau.threshold_mode
cumulative_delta	cumulative	Shorter, clearer

Backward Compatibility

• Old parameter names still work — they emit a DeprecationWarning and forward the value to the new name
• If both old and new names are passed simultaneously, a ValueError is raised with a clear message
• All existing code using old names will continue to work without changes

Changes

ignite/handlers/early_stopping.py

• Added new parameter names as primary, old names kept as deprecated Optional kwargs
• warnings.warn() with DeprecationWarning for each old name
• ValueError on conflict (both old + new specified)
• Internal attributes and __call__ use new names
• Updated docstrings with versionchanged:: 0.5.5 note

tests/ignite/handlers/test_early_stopping.py

• Updated existing tests to use new parameter names
• Added 8 new test cases:
  • test_args_validation_deprecated_names — validation through deprecated params
  • test_deprecated_min_delta — DeprecationWarning + value forwarding
  • test_deprecated_cumulative_delta — same for cumulative_delta
  • test_deprecated_min_delta_mode — same for min_delta_mode
  • test_deprecated_and_new_param_conflict — ValueError when both passed
  • test_backward_compat_deprecated_params_still_work — end-to-end with old names
  • test_backward_compat_cumulative_delta — end-to-end with deprecated cumulative
  • test_backward_compat_min_delta_mode — end-to-end with deprecated threshold_mode

Test Plan

• All 28 non-distributed EarlyStopping tests pass
• Old parameter names emit DeprecationWarning
• New parameter names work correctly
• Conflict detection works (ValueError)
• No regression in existing behavior

[joemunene-by]

Thanks for letting me know @ramyars466! I can see your PR #3619 already covers this — glad it's approved and ready to go. Closing this one in favor of yours.

[ramyars466]

Thank you, @joemunene-by! I really appreciate your understanding and support. Looking forward to seeing your future contributions to the project!